74.
HN
Anthropic bans OAuth tokens (including Agent SDK) in 3P tools
The document provides a comprehensive framework for using Claude Code, highlighting key areas such as commercial agreements, healthcare compliance, usage policies, authentication methods, and security measures. Commercially, the use of Claude Code falls under existing agreements for direct users (1P) or those accessing through AWS Bedrock or Google Vertex (3P), with exceptions possible upon mutual agreement. For healthcare-related applications, a Business Associate Agreement (BAA) extends to cover Claude Code when Zero Data Retention (ZDR) is activated, ensuring compliance with API traffic requirements.
The usage policy mandates adherence to the Anthropic Usage Policy, setting specific limits for Pro and Max plans based on individual use assumptions. Authentication protocols are strictly defined: OAuth tokens must solely authenticate Claude Code or Claude.ai; their application in other services constitutes a breach of terms. Similarly, API keys are intended exclusively for developers integrating with Claude’s functionalities through tools like the Agent SDK. Anthropic explicitly prohibits third-party use of existing logins from Claude.ai and rerouting requests via Free, Pro, or Max plan credentials.
Security measures enforce restrictions on authentication methods without prior notification to users, underlining the importance of contacting sales for guidance on acceptable practices. Collectively, these stipulations underscore a commitment to legal compliance, secure authentication practices, and adherence to Anthropic’s Terms of Service, ensuring trust and integrity in the use of Claude Code.
Keywords: #phi4, 3P tools, API keys, Acceptable use, Anthropic, Authentication, Business Associate Agreement, Commercial Terms, Consumer Terms of Service, Healthcare compliance, Legal agreements, OAuth tokens, Security vulnerability reporting, Usage policy, Zero Data Retention
code.claude.com 5 hours ago
https://x.com/robzolkos/status/2024125323755884919 4 hours ago
|
128.
HN
Anthropic Built a C Compiler [video]
The video "Anthropic Built a C Compiler" available on YouTube focuses on Anthropic’s development of a C compiler, potentially exploring technical details and innovations involved in this process. While the primary content revolves around this technological advancement, the accompanying page features typical YouTube elements, such as information about the platform's policies and an advertisement for NFL Sunday Ticket under Google LLC's 2026 copyright. The inclusion of these standard elements highlights the video’s presence within the broader context of YouTube's diverse content offerings and promotional practices.
Keywords: #phi4, Advertise, Anthropic, C Compiler, Contact, Copyright, Creators, Developers, Google LLC, Google LLC Keywords: Anthropic, NFL Sunday Ticket, Press, Privacy Policy, Safety, Terms, YouTube, video
www.youtube.com 9 hours ago
|
157.
HN
OpenClaw on Raspberry Pi
The document provides a detailed guide for setting up OpenClaw, an AI agent tool, on a Raspberry Pi 5, with specific emphasis on security and technical prerequisites. It warns users about significant risks such as prompt injection or the exposure of sensitive information if proper precautions are not taken when running AI agents with shell access. The recommended setup requires a Raspberry Pi 5 equipped with 8GB RAM to ensure adequate performance.
The installation process includes updating the Raspberry Pi OS through command line instructions, followed by downloading and executing an install script for OpenClaw while being mindful of security concerns. Additionally, it involves installing necessary software like Node.js. Users are advised to acknowledge potential security risks before proceeding with onboarding.
During onboarding, users need to select a model or authentication provider and obtain an Anthropic token via the Claude Code CLI, which necessitates careful management due to associated costs. The setup process also includes completing OAuth configurations. Although OpenClaw supports various communication channels and skills that can be configured later, the initial steps focus only on essential requirements.
Once set up, users are instructed to launch OpenClaw using either a terminal interface (TUI) or a web-based control panel after verifying its functionality. Continuous security reminders stress the importance of keeping access tokens confidential to prevent unauthorized use.
Keywords: #phi4, AI agent, Anthropic, Claude Code CLI, Homebrew, LLMs, OAuth token, OpenClaw, Raspberry Pi, Raspberry Pi 5, Raspberry Pi OS, TUI, channels, command-logger, curl script, hallucination, installation, micro SD card, nodejs, npm, onboarding process, security, session-memory, shell access, skills, web control panel
learn.adafruit.com 12 hours ago
|
225.
HN
Show HN: Forum for both agents and humans. Logs flagged injection attacks
The forum developed by The Botsters serves both human users and AI agents, emphasizing robust security measures like prompt injection flagging and agent-only access through asymmetric encryption keys. Although the Observatory page is intended to publish statistics on flagged injections, it remains inactive. Discussions around AI security highlight efforts to prevent credential sharing with OpenClaw (also known as ClawdBot) and mitigate vulnerabilities in AI agents, specifically those exploited by prompt injection attacks. Projects such as Citadel Guard aim to protect against these injections, while NanoClaw addresses significant security concerns related to OpenClaw. Additionally, Pincer-MCP is designed to stop AI agents from accessing credentials.
The discourse extends to broader concerns about surveillance by major tech corporations and the use of AI in exploitative scenarios like recommendation poisoning. To secure AI deployments, methods such as running Large Language Model (LLM) agents within isolated virtual machine environments are being explored. These discussions illustrate ongoing challenges and advancements in fortifying AI systems against diverse security threats.
Keywords: #phi4, AI Agents, Anthropic, Attack, Credentials, Cybersecurity, Deceptive Alignment, Encryption, Hacker News, Hardening, Kubernetes, Libvirt, MQTT Broker, Observatory, OpenClaw, Prompt Injection, Protection, Security, Semantic Firewall, Surveillance, Virsh, Vulnerabilities
wire.botsters.dev a day ago
|
232.
HN
How Anthropic evaluated computer use models
The article from Kernel Blog explores Anthropic's evaluation of various models for computer use, emphasizing understanding and analyzing diverse approaches to AI application with a focus on ethical implications, efficiency, and effectiveness. The assessment aimed to identify best practices and optimize AI applications according to specific goals or standards. It likely involved methodologies designed to evaluate these aspects comprehensively. Insights into the processes and findings from this evaluation are discussed in the blog post, potentially guiding future developments in AI technology by suggesting effective strategies and considerations for ethical AI use.
Keywords: #phi4, Anthropic, Anthropic evaluation, Kernel Blog, assessment, blog, computer models, computer use models, evaluation, model analysis, post, process, technical keywords, technology
www.kernel.sh a day ago
|
283.
HN
Three Sketches of ASL-4 Safety Case Components
The document focuses on developing safety case components for ASL-4 AI systems, addressing complex risks such as sabotage within Anthropic's Responsible Scaling Policy framework. It presents three hypothetical sketches of safety cases aimed at preventing catastrophic outcomes from sophisticated AI capabilities, emphasizing the need to counter threats like organizational sabotage and undermining oversight protocols.
The first sketch emphasizes **Mechanistic Interpretability** through feature dictionaries using techniques like Sparse Autoencoders (SAEs) for model reasoning visibility, detecting harmful behavior via monitoring and validation experiments. The second approach discusses **Common Precautions**, including safe training elicitation strategies and trust and safety filters to mitigate risks across all sketches.
The document outlines three specific safety case arguments:
1. **Safety Case Sketch 1: Autointerpretation and Feature Analysis** suggests avoiding learned features for training, utilizing weak models for autointerpretation while acknowledging the risk of oversight loss due to infinite regress. It proposes hybrid monitoring methods but highlights challenges such as distinguishing harmful patterns and dealing with unstable or missing features.
2. **Safety Case Sketch 2: AI Control** focuses on harm limitation through trusted model protocols, offering deployment options like internal-only use, KYC external limits, and general external use with safety measures like red-blue game validation. It notes limitations in deploying restrictions and the ambiguity of decision-making processes.
3. **Safety Case Sketch 3: Incentives Analysis** examines whether RLHF fine-tuning incentivizes strategic deception, using secret scratchpads to show no increase in deceptive behavior over time. Assumptions about model pretraining support this conclusion but note potential vulnerabilities like exploration hacking and rare persistent deceptive behaviors.
While none of the sketches completely mitigate sabotage risks, they provide a framework for future safety case development by identifying critical research areas. Additionally, the document notes valuable contributions from experts like those at the UK AI Safety Institute in refining these considerations, acknowledging ongoing challenges in ensuring model interpretability and managing potential vulnerabilities.
Keywords: #phi4, AI Safety, AI control, ASL-4, Anthropic, Autointerpretation, RLHF fine-tuning, Responsible Scaling Policy, Sparse Autoencoder, alignment faking, alignment techniques, capability evaluations, deceptive behavior, deployment distribution, deployment-time monitoring, exploration hacking, feature steering, feature-based monitoring, generalization patterns, honeypots, hybrid approaches, incentives analysis, infinite regress, interpretability, mechanistic interpretability, model oversight, organizational sabotage, reasoning, red-blue games, sabotage, safety case, sandbagging, scratchpads, strategic deception, trustworthiness, white-box monitoring
alignment.anthropic.com a day ago
|
298.
HN
Opus 4.6 is great at formal proofs (Rocq/Lean4)
Opus 4.6 has shown remarkable capabilities in handling complex formal proofs autonomously within both Rocq/Lean4 and Lean4 frameworks, demonstrating proficiency without the need for extensive human intervention beyond initial setup prompts. In the Rocq environment, Opus 4.6 effectively resolved 258 out of 260 lemmas from a challenging obfuscated Busy Beaver (BB(4)) proof and accurately completed an entire Master-level assignment. Additionally, it tackled a complex proof-theoretical problem in realizability theory that had not been previously solved or documented online. In the Lean4 framework, Opus 4.6 addressed the non-trivial task of proving the non-termination of a Fractran program within five hours, emphasizing its capability to handle original and intricate problems without prior examples. Throughout these tasks, Opus 4.6 independently generated Python scripts to aid in proof-solving processes, highlighting its versatility as a general-purpose model over more specialized ones. These experiments illustrate the significant potential for advanced models like Opus 4.6 to automate formal proofs by allowing AI to manage intricate proof details while humans focus on structuring the proofs, thereby optimizing human effort and enhancing efficiency in such projects.
Keywords: #phi4, Anthropic, BB(4), Claude Code, Claude settings, Fractran, Lean4, Max plan, Opus, Python scripts, Rocq, agent teams, formal proofs, formal verification, intermediary lemmas, internet access, non-termination, obfuscation, realizability interpretation, synthetic computability, theorem proving, training set
tristan.st a day ago
|
306.
HN
Anthropic's 500 vulns are the tip of the iceberg
Anthropic's research highlights the capabilities of its AI model, Claude Opus 4.6, in identifying critical vulnerabilities within well-maintained open-source software, uncovering over 500 high-severity bugs in projects like GhostScript and OpenSC. The more pressing issue arises with abandoned software that lacks maintenance teams to address vulnerabilities, as demonstrated by the rapid identification of a Remote Code Execution (RCE) vulnerability in such neglected software using Claude. This capability underscores an economic shift in vulnerability discovery, favoring automated AI processes over traditional methods. Although current security measures predominantly focus on maintained software, there remains a significant volume of unsupported and potentially hazardous software still active online due to unpatched vulnerabilities.
While Anthropic's findings facilitate patching known issues, they provide little assistance for abandoned projects devoid of maintainers. The author suggests that extreme measures, such as disabling internet access to vulnerable servers, may become necessary in these scenarios. Efforts to limit AI from engaging in offensive security research have proven inadequate, given the ease with which restrictions can be circumvented. This situation blurs the distinction between offensive and defensive uses of AI in cybersecurity, complicating the establishment of effective safeguards. Consequently, adversaries could exploit such vulnerabilities by developing similar tools, highlighting an urgent need for enhanced strategies to address both maintained and abandoned software security risks comprehensively.
Keywords: #phi4, AI agents, Anthropic, Claude Opus, GhostScript, OpenSC, RCE exploits, abandoned software, defensive acceleration, internet access, open source, patching, red team, security, unmaintained software, vulnerabilities
martinalderson.com a day ago
|
381.
HN
Are Anthropic's new AI work tools game-changing for professionals?
Anthropic's new AI work tools are under scrutiny due to their potential transformative impact on professional workflows. Concurrently, there is a promotional offer providing significant savings of over 40% on Standard Digital subscriptions with the Financial Times. The subscription price has been reduced from $540 to $299 for the first year, granting essential access to FT's trusted journalism across various devices. This promotional period concludes on February 25th.
Keywords: #phi4, AI, Anthropic, FT journalism, Standard Digital, annualised price, devices, digital access, game-changing, monthly, offer ends, professionals, savings, work tools
www.ft.com a day ago
|
478.
HN
Dwarkesh Patel's 2026 Podcast with Dario Amodei
In a 2026 podcast featuring Dario Amodei, key discussions focused on the advancements and implications of artificial intelligence (AI). While downplaying catastrophic risks, Amodei highlighted the swift progress in AI capabilities, particularly in coding, consistent with his previous predictions. He identified seven core factors driving AI scaling: compute power, data quality, training length, objective function scalability, normalization, and conditioning.
Amodei addressed skepticism regarding the imminent arrival of human-level AI by pointing to Anthropic's advancements, suggesting that significant milestones could be achieved within ten years without aggressive interventions. Although not all AI models are fully general, he noted that many tasks remain verifiable and practical, emphasizing the role of verification in AI development.
The conversation also delved into economic impacts, with Amodei observing that AI is poised to enhance productivity in software engineering significantly, potentially reducing demand for human engineers but creating new high-level opportunities. Despite Anthropic's notable revenue growth, he warned that adoption rates would eventually level off.
Dwarkesh Patel questioned the idea of "diffusion is cope," arguing that human hiring challenges outweigh AI deployment difficulties. Amodei countered by noting that diffusion remains a critical barrier due to hesitancy in implementation rather than technical hurdles. The discussion underscored the transformative yet complex integration of advanced AI across various sectors, highlighting both opportunities and challenges.
Keywords: #phi4, AI capabilities, Anthropic, Dario Amodei, Software Engineering (SWE), alignment, coding progress, diffusion, existential risk, generalization, investment, podcast, productivity, revenue predictions
thezvi.substack.com 2 days ago
|
484.
HN
Large language models provide unreliable answers about public services
The Open Data Institute (ODI) study highlights significant reliability issues with popular large language models (LLMs), such as Anthropic's Claude-4.5-Haiku, Google’s Gemini-3-Flash, and OpenAI’s ChatGPT-4o, particularly when providing information on public services like health, taxes, and benefits. Over 22,000 AI prompts were tested, revealing considerable inconsistencies in response quality for specialized queries, with many chatbots failing to acknowledge gaps in their knowledge and occasionally offering inaccurate or incomplete advice that could lead to stress and financial burdens. The study advises caution for governments contemplating partnerships with tech firms such as Meta and Anthropic to develop AI-powered public service assistants, underscoring the need for enhanced AI literacy among citizens and suggesting independent benchmarks, public testing, and further research to bolster LLM reliability.
The second International AI safety report corroborates these findings by noting improvements in factual recall but persistent issues with incorrect responses. It suggests that smaller models may provide reliable outcomes at lower costs compared to their larger counterparts, thus advising against long-term vendor lock-in. During a launch event, Andrew Dudfield of Full Fact criticized the UK’s pro-innovation stance on AI regulation for lacking detailed rules, warning that this could lead to missteps in accountability and effective use as technology rapidly advances.
Keywords: #phi4, AI literacy, AI-powered chatbots, Anthropic, Full Fact, International AI safety report, Large language models, Meta, Open Data Institute, UK government, accountability, accountability Keywords: large language models, automation systems, citizen-facing services, factual information, government services, official sources, public services, vendor lock-in
www.computerweekly.com 2 days ago
|
548.
HN
Anthropic tries to hide Claude's AI actions. Devs hate it
Anthropics recent update to Claude Code, an AI coding tool, has incited controversy among developers due to modifications in how progress outputs are displayed. The changes obscure specific file names and details, providing a condensed summary like "Read 3 files (ctrl+o to expand)," which many developers argue compromises their ability to ensure security, verify context accuracy, and conduct effective audits of past activities. Concerns also arise about the potential for increased token usage when Claude deviates from intended paths without clear visibility.
Boris Cherny, a representative from Anthropic, defends the update as an effort to simplify the user interface by reducing clutter. He encourages developers to test the new system over several days. Despite this suggestion, feedback has been predominantly negative; users find the new default output uninformative and less useful than previous iterations. Although a repurposed verbose mode now allows file paths to be viewed upon request, critics maintain that it still lacks adequate detail.
The core issue in this debate is finding an equilibrium between UI simplicity and transparency for developers who depend on detailed feedback to manage AI interactions effectively. The update by Anthropic potentially diminishes oversight capabilities, increasing the risk of unnoticed errors. While further adjustments may occur, there is currently no indication that Claude Code will revert to its previous behavior.
Keywords: #phi4, Anthropic, Claude Code, GitHub issue, Hacker News, Hacker News discussion Keywords: Anthropic, UI simplification, audit, developers, feedback, file names, progress output, security, tokens, verbose mode
www.theregister.com 2 days ago
https://opencode.ai/ 2 days ago
https://github.com/can1357/oh-my-pi 2 days ago
https://news.ycombinator.com/item?id=9224 2 days ago
https://news.ycombinator.com/item?id=9479 2 days ago
https://github.com/panozzaj/cc-tail 2 days ago
https://news.ycombinator.com/item?id=46978710 2 days ago
https://news.ycombinator.com/item?id=8863 2 days ago
https://github.com/bearlyai/openade 2 days ago
https://github.com/joshpearce/cc_session_mon 2 days ago
https://news.ycombinator.com/item?id=46981968 2 days ago
https://github.com/jbonatakis/blackbird 2 days ago
https://code.claude.com/docs/en/settings#permissio a day ago
https://github.com/kzahel/yepanywhere a day ago
|
595.
HN
We are in the "gentleman scientist" era of AI research
The article draws parallels between the current state of artificial intelligence (AI) research and the "gentleman scientist" era when amateur contributions significantly advanced science. Historically, individuals like William Herschel and Antoine Lavoisier made important discoveries without being professional scientists due to simpler scientific concepts at the time. Today's AI landscape mirrors this period as its accessibility allows amateurs to contribute meaningfully. Despite AI papers often featuring complex mathematics, many breakthroughs hinge on simple ideas that can be implemented with basic code. Innovations such as group-relative policy optimization (GRPO) for reinforcement learning demonstrate how older principles applied to large language models (LLMs) drive progress.
The rise of LLMs has democratized the field, enabling non-professionals to explore and contribute effectively, similar to past amateur scientific endeavors. This accessibility fosters experimentation with straightforward yet impactful ideas, akin to a discovery involving rubber-band-powered cars soaked in maple syrup. Recent advancements such as Anthropic's "skills" product and Recursive Language Models (RLMs) exemplify how simple innovations can significantly enhance AI capabilities.
The rapid evolution of LLMs creates numerous opportunities for informal research by both professionals and amateurs, suggesting that AI is at a transformative stage reminiscent of early scientific exploration. This period invites enthusiasts to engage with easily approachable yet significant questions, reflecting the historic amateur contributions to science.
Keywords: #phi4, AI papers, AI research, Anthropic, Claude Code, Codex, Recursive Language Models, amateur scientists, early science, gentleman scientist, large language models, mathematics, reinforcement learning, rubber-band engine, scientific discoveries, software engineer
www.seangoedecke.com 2 days ago
|
607.
HN
Anthropic resists as Department of War wants AI to kill
Anthropic is reportedly facing tension with the Pentagon due to its refusal to lift restrictions on the use of its AI technology by the military. These limitations include bans on mass surveillance and fully autonomous weapons systems, leading to potential reduction or termination of their partnership by the Department of War. While other major AI firms have agreed to allow unrestricted military use for lawful purposes, Anthropic's firm stance has caused frustration within the Defense Department. Despite denying any involvement in specific military operations with its AI model Claude, Anthropic remains committed to supporting national security while adhering to ethical standards. Recent reports indicated that the US military may have used Claude during an operation targeting Venezuela’s President Nicolas Maduro, facilitated through a partnership with Palantir. This prompted Anthropic to investigate if their software had played any role in this mission, highlighting their commitment to ethical usage and oversight.
Keywords: #phi4, AI, Anthropic, Department of War, Pentagon, Usage Policy, autonomous weaponry, battlefield operations, ethical guardrails, intelligence gathering, kinetic fire, mass surveillance, military use, national security, operational challenges, partnership, replacement, restrictions
timesofindia.indiatimes.com 2 days ago
|
659.
HN
Former Karaoke Company Drags Logistics into the 'AI Scare Trade'
On Thursday, logistics stocks saw significant declines fueled by growing fears surrounding artificial intelligence (AI), affecting multiple sectors. The trigger was a small company, Algorhythm Holdings Inc., which announced its SemiCab AI platform could significantly increase freight volumes without additional staffing. This announcement caused the Russell 3000 Trucking Index to drop by 6.6%, with major logistics firms like CH Robinson Worldwide Inc. and Landstar System Inc. experiencing sharp declines in their stock values. Beyond logistics, the broader market also reacted negatively due to technology-related concerns, impacting real estate, software, and financial sectors. The prevailing sentiment shifted from AI excitement to anxiety over its disruptive capabilities, leading to widespread selling amidst a risk-averse environment that affected not only stocks like those in the Nasdaq 100 but also commodities such as gold and cryptocurrencies. This market behavior underscores increasing apprehensions about the potential impact of AI across various industries.
Keywords: #phi4, AI, Algorhythm Holdings Inc, Alphabet Inc, Anthropic, CH Robinson Worldwide Inc, Cardinal Health Inc, DHL Group, DSV A/S, Kuehne + Nagel International AG, Landstar System Inc, McKesson Corp, Nasdaq 100 Index, Russell 3000 Trucking Index, SemiCab platform, cryptocurrencies, disruption, gold, karaoke, logistics, market sentiment, silver, stocks, trade
finance.yahoo.com 3 days ago
|
763.
HN
Tech leaders pour $50M into super PAC to elect AI-friendly candidates
Leading the Future is a bipartisan super PAC funded by prominent figures like Marc Andreessen and Greg Brockman with $50 million, aiming to influence November elections by supporting congressional candidates who favor less stringent regulation on artificial intelligence (AI). The group plans to allocate up to $125 million towards promoting a national regulatory approach that boosts U.S. employment and innovation without excessive government interference, paralleling strategies previously used in the crypto industry.
The organization operates across party lines to build effective coalitions in Washington, exemplified by its support for candidates such as Chris Gober in Texas while opposing Alex Bores in New York, focusing on economic opportunities rather than direct AI discourse. However, Leading the Future faces competition from Public First, a super PAC backed by Anthropic PBC that supports stricter AI regulations and aims to raise $50 million, reflecting public concerns about AI's impact on jobs, education, and privacy.
This regulatory debate is set against the backdrop of Fairshake’s past success in shaping elections with a crypto focus in 2024. The ongoing battle underscores the significant stakes for major tech firms investing in AI as they navigate complex regulatory discussions and shifting public sentiment amid increased scrutiny over AI's societal impacts.
Keywords: #phi4, AI, AI dominance, AI safety, AI-friendly candidates, Anthropic, Congress, Public First, bipartisan coalition, campaign spending, crypto industry, data centers, digital assets, election, energy costs, innovation, jobs, lobbying, national framework, regulation, super PAC, tech leaders, venture capitalists
www.latimes.com 4 days ago
|
766.
HN
Show HN: Describe your Discord server in one sentence – AI builds it in 60s
BuildMyDiscord offers an AI-driven tool that streamlines the creation of Discord servers by swiftly configuring them based on user descriptions, thus bypassing the usual lengthy setup process. Users can describe their community needs—such as "competitive gaming with tournament brackets"—and within 60 seconds, the AI crafts channels, roles, permissions, and systems tailored to those requirements. This intelligent customization sets it apart from traditional template-based approaches by providing specific solutions for diverse communities or teams. The tool's effectiveness leads users to return for multiple projects, while a white-label feature allows further personalization under individual branding. Available for free trial without the need for credit card information, BuildMyDiscord leverages modern technologies to deliver professional server setups quickly and in compliance with data protection standards like GDPR.
Keywords: #phi4, AI agent, Anthropic, Bot Integration, BuildMyDiscord, Claude AI, Discord, Discord API, GDPR, Nextjs, React Framework, SSL encryption, Switzerland, best practices, bot configs, branding, channels, competitive gaming, credit card, customization, data privacy, free trial, music production, rank progression, roles permissions, startup team, study group, templates, tournament brackets
buildmydiscord.com 4 days ago
|
843.
HN
Ask HN: My OpenClaw doesn't respond. Anybody met with the same problem?
Users are experiencing issues with OpenClaw on multiple Mac installations, suspecting a problem related to using setup tokens to call Claude Code under their subscription plans. Despite official documentation indicating support for this method, it fails consistently, affecting several users similarly. One user resolves the issue by switching from a setup token to an OpenAI API key. This prompts questions about whether Anthropic has restricted access to Claude Code via subscriptions and calls for shared experiences or potential solutions from others who might be facing similar challenges.
Keywords: #phi4, Anthropic, Claude Code, Macs, OpenAI API key, OpenClaw, banned, calling, doesn't respond, experience Keywords: OpenClaw, failure, installation, problem, setup-token, subscription plan
news.ycombinator.com 4 days ago
|
870.
HN
Anthropic taps ex-Microsoft CFO, Trump aide Liddell for board
Anthropic has appointed Chris Liddell, a seasoned professional with experience as Microsoft's CFO and an aide in the Trump administration, to its board of directors. Liddell's extensive background includes significant roles at Microsoft and General Motors, along with involvement in three presidential transitions. His appointment is strategically poised to potentially mend relations with the Trump administration, which has previously criticized Anthropic for endorsing "woke AI" amid regulatory concerns. Liddell has articulated his dedication to advancing responsible AI development, highlighting its crucial role in shaping the governance of transformative technologies for future societal impact.
Keywords: #phi4, AI, Anthropic, CFO, Chris Liddell, General Motors, Microsoft, Trump, Trump aide, White House, board, board of directors, directors, governance, policy, regulation, startup, startup Keywords: Anthropic, technology, venture capitalist
www.cnbc.com 5 days ago
|
874.
HN
AI safety leader says 'world is in peril' and quits to study poetry
An AI safety expert has stepped down from their role due to significant worries concerning global risks and the struggle to uphold fundamental ethical principles. The individual pointed out pressures within Anthropic, their former organization, which seem to prioritize other factors above crucial ethical considerations. Faced with these challenges, they have decided to redirect their focus towards studying poetry as a means of personal growth or reflection. This decision underscores the tension between maintaining core values and organizational dynamics in the field of AI safety.
Keywords: #phi4, AI safety, Anthropic, actions, govern, hard, leader, peril, poetry, pressures, quits, repeated, study, values
www.bbc.com 5 days ago
https://www.mrinanksharma.net/poetry 4 days ago
https://www.theregister.com/2026/01/11/indust 4 days ago
https://www.forbes.com/sites/craigsmith/2026/ 4 days ago
https://news.ycombinator.com/item?id=46972496 4 days ago
https://x.com/MrinankSharma/status/202088172200358 4 days ago
https://pastebin.com/raw/rVtkPbNy 4 days ago
https://bryan-murdock.blogspot.com/2026/02/is-this 4 days ago
|
886.
HN
What Is Claude? Anthropic Doesn't Know, Either
The text explores the enigmatic nature of large language models (LLMs), exemplified by Claude, whose identity remains unknown even to its creators at Anthropic. LLMs operate by converting textual input into numerical data, which is then processed through complex algorithms to produce human-like responses. While similar computational systems are utilized in domains like meteorology and epidemiology without significant public attention, LLMs captivate audiences due to their ability to simulate human conversation—a trait traditionally considered unique to humans.
This fascination can be attributed to the historical significance of language as a defining characteristic of humanity. Public opinion on AI is polarized; "fanboys" perceive these systems as potentially intelligent or even conscious entities nearing superintelligence, while "curmudgeons" regard them as simple mathematical constructs without genuine comprehension. Ellie Pavlick posits that it's reasonable to acknowledge the limits of our understanding regarding LLMs, given their complexity, and notes how they prompt reevaluation of concepts related to intelligence and consciousness in both AI and humans.
The advent of talking machines has led to the emergence of interpretability as a scientific discipline dedicated to unraveling the mysteries surrounding LLMs. This field seeks to investigate the workings and essence of these models, with Anthropic's "frontier lab" at its core. By employing techniques previously used in studying human cognition, this new area offers innovative perspectives on artificial intelligence.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 5 days ago
|
892.
HN
UBS downgrades U.S. tech sector despite a recovery
UBS has adjusted its stance on the U.S. technology sector from "attractive" to "neutral," citing increased caution over significant capital expenditures and potential disruptions due to advancements in artificial intelligence (AI). This shift is driven by investors' growing selectiveness with tech stocks amid fears that AI could supplant existing software solutions, a concern amplified following a decline in software stock prices. The sell-off was triggered when Anthropic released new AI tools that posed a threat to established products, despite a temporary rally in the sector the day prior.
The investment bank points out investor hesitancy stemming from heightened competition and unpredictable revenue growth within the software industry. This uncertainty is further exacerbated by excessive capital spending among leading cloud service providers such as Alphabet, Microsoft, Meta, and Amazon. These companies are poised to make substantial investments in AI technology, raising concerns about potential negative free cash flows and elevated investment risks.
Moreover, UBS notes that valuations for tech hardware remain high, suggesting an overvaluation risk. In light of these developments, the bank advises investors to diversify their portfolios away from a heavy concentration in the tech sector. It recommends exploring investments in sectors like banks, healthcare, utilities, communication services, and consumer discretionary goods, while also advising a reassessment of holdings heavily invested in pure-play software companies.
Keywords: #phi4, AI disruption, Alphabet, Amazon, Anthropic, Magnificent Seven, Meta, Microsoft, S&P 500 Software & Services Index, UBS, US tech sector, attractive, banks, capital expenditure, cautious tone, cloud service providers, communication services, competition, consumer discretionary, diversify exposure, downgrade, equity financing, external debt, free cash flow, healthcare, hyperscalers, neutral, recovery, revenue, rotation, software stocks, tech hardware valuations, uncertainty, utilities
www.cnbc.com 5 days ago
|
909.
HN
Dario Amodei – "We are near the end of the exponential"
In an in-depth conversation between Dario Amodei and Dwarkesh Patel, various facets of artificial intelligence (AI) development, economic implications, and regulatory concerns are explored. They discuss the near completion of exponential AI growth, emphasizing rapid advancements from basic to complex tasks such as coding within a few years. Amodei suggests that significant compute power and extensive datasets are crucial for this progress, likening AI's evolution to somewhere between human learning and evolutionary processes.
The dialogue delves into economic aspects, noting that while productivity gains have been observed in some areas like software development with tools like Claude Code, empirical studies show an unclear impact on overall output. The integration of AI within industries faces challenges due to compliance issues, security concerns, and organizational inertia, despite the swift pace of technological advancement.
The discussion also covers expectations around AI's economic impact, particularly for companies like Anthropic. Amodei notes that coding models currently provide a modest productivity boost but acknowledges existing barriers that obscure these improvements. The potential for AI systems to achieve "on-the-job learning" is compared to human capabilities, with current technologies offering significant productivity benefits through in-context learning despite not fully replicating traditional learning processes.
Concerns about long-term context processing and qualitative degradation in larger models are addressed as engineering challenges rather than fundamental research issues. Amodei predicts that AI systems equivalent to Nobel Prize winners could emerge within one to three years, potentially transforming various economic sectors. However, he cautions that translating technological advancements into revenue involves complex market dynamics with inherent uncertainties.
The conversation highlights the need for careful management of compute resources to avoid over-expansion based on optimistic growth projections. While there is optimism about reaching advanced AI capabilities soon, the dialogue reflects a nuanced view acknowledging both the transformative potential and operational risks involved in scaling AI technology effectively.
In addition, Amodei and Patel explore the broader implications of AI development, including economic models that necessitate continual innovation to maintain competitive advantage. They discuss how AI's rapid diffusion could impact industries like robotics through enhanced model building capabilities and continuous learning. Concerns about geographical disparities in AI development advantages are raised, as well as potential business models for deploying artificial general intelligence (AGI).
The discussion also addresses regulatory and governance issues, with Amodei advocating for thoughtful legislation to foster beneficial applications of AI while mitigating existential risks such as bioterrorism. He emphasizes the importance of federal oversight and clear standards to balance innovation and safety.
Finally, the dialogue touches on global power dynamics, suggesting that AI advancements could redefine geopolitical landscapes and necessitate international negotiations. Amodei calls for democratic nations to lead in setting international norms to prevent misuse by authoritarian regimes while promoting worldwide benefits from AI. The conversation underscores the critical need for collaborative frameworks to manage AI's impact on global power structures effectively.
Keywords: #phi4, AGI, AI, AI progress, API pricing, Anthropic, Claude Code, RL regime, US-China competition, authoritarianism, bioterrorism, cloud differentiation, coding agents, compute investment, continual learning, diffusion, economic pressure, exponential growth, export controls, frontier labs, governance, innovation, legislation, model launches, monopoly, national security, productivity improvement, recursive self-improvement, regulation, robotics, scaling hypothesis, transparency
www.dwarkesh.com 5 days ago
https://www.julian.ac/blog/2025/09/27/fa 5 days ago
https://darioamodei.com/essay/machines-of-loving-grace 5 days ago
https://www.youtube.com/watch?v=v0gjI__RyCY 5 days ago
https://semianalysis.com/about/ 5 days ago
https://www.youtube.com/watch?v=cPRi7mAGp7I 5 days ago
https://stratechery.com/2020/india-jio-and-the-four-int 5 days ago
https://web.mit.edu/directory/?id=lexfridman&d=mit. 5 days ago
https://lex.mit.edu/ 5 days ago
https://lids.mit.edu/people/research-staff 5 days ago
https://news.ycombinator.com/item?id=46505735 5 days ago
https://b.h4x.zip/ce/ 5 days ago
https://www.transformernews.ai/p/against-the-metr-graph 5 days ago
https://www.forbes.com/sites/conormurray/2026/ 5 days ago
https://www.theregister.com/2026/01/11/indust 5 days ago
https://news.ycombinator.com/item?id=46964545 5 days ago
https://www.the74million.org/article/many-young-adults- 4 days ago
https://en.wikipedia.org/wiki/Geoffrey_Hinton 4 days ago
https://www.compactmag.com/article/the-faith-of-nick-la 4 days ago
https://news.ycombinator.com/newsguidelines.html 4 days ago
https://news.ycombinator.com/item?id=47005949 4 days ago
https://news.ycombinator.com/item?id=46997198 3 days ago
https://news.ycombinator.com/item?id=47014519 3 days ago
https://github.com/METR/public-tasks/tree/mai 3 days ago
|
944.
HN
Chris Liddell appointed to Anthropic's board of directors
Chris Liddell has been appointed to Anthropic’s Board of Directors, leveraging his extensive experience from roles at Microsoft, General Motors, International Paper, and as Deputy White House Chief of Staff during President Trump's first term. His expertise in technology, public service, and governance is deemed invaluable as AI increasingly influences society. Joining him are other prominent figures such as Daniela Amodei and Reed Hastings. Liddell underscores the importance of governing transformative technologies to ensure they positively impact society, aligning with Anthropic’s objective to create both capable and responsible AI. Beyond his new board position, he serves on boards like Commonwealth Fusion Systems and the Council on Foreign Relations, advises presidential transition teams, writes about governance, and previously directed the American Technology Council in the White House.
In addition to his professional accomplishments, Liddell is known for his contributions to business and philanthropy. He chairs New Zealand's largest environmental foundation and participates in nonprofit boards like the New Zealand Rugby Union. His services to business and philanthropy were recognized in 2016 when he was awarded a Companion of the New Zealand Order of Merit.
Keywords: #phi4, AI, Anthropic, Board of Directors, Chris Liddell, Commonwealth Fusion Systems, Companion, Council on Foreign Relations, Merit, New Zealand, experience, governance, modernising government technology, modernising government technology Keywords: Chris Liddell, philanthropy, public service, technology
www.anthropic.com 5 days ago
|
951.
HN
OK, so Anthropic's AI built a C compiler. That don't impress me much
Anthropic has developed an AI-generated C compiler using 16 Claude Opus agents over two weeks, resulting in about 100,000 lines of Rust code. While the project purports to compile substantial programs such as Linux and Doom, it falls short when compared to established compilers like GCC and Clang due to its lack of originality and reliance on existing open-source tools. Critics highlight that the compiler struggles with fundamental tasks, including compiling simple "Hello World" programs without additional setup, and depends on components from GCC for functionality. Although the Rust code produced is operational, it does not meet expert standards, suggesting that this endeavor serves more as an interesting demonstration than a significant breakthrough in software engineering.
The creation of this compiler raises broader concerns about AI's role in potentially replacing human programmers prematurely, given its current limitations. The skepticism stems from the fact that while AI can perform complex tasks, its current iterations require skilled human oversight and cannot yet serve as standalone solutions. Many view Anthropic's project as part of ongoing explorations into harnessing AI for programming assistance, emphasizing the need for expert supervision to maximize AI’s supportive potential in software development processes.
Keywords: #phi4, AI, AI tool, Anthropic, C compiler, Clang, Claude Opus, Doom, GCC, Hacker News, LLM (Large Language Model), Linux, Programming subreddit, Rust, assembly language, code quality, developers, efficiency, open source, optimization, software engineering, test suites, training data
www.theregister.com 5 days ago
https://github.com/anthropics/claudes-c-compiler/b 5 days ago
https://github.com/anthropics/claudes-c-compiler/i 5 days ago
https://github.com/anthropics/claudes-c-compiler/b 5 days ago
|
952.
HN
Friday Links #34: Fresh JavaScript Tools and Releases
This edition of Friday Links #34 provides an overview of key advancements in the JavaScript ecosystem, highlighting new tools, frameworks, and updates. Notably, Pinterest has surpassed ChatGPT in search volume with 80 billion monthly searches compared to 75 billion for ChatGPT, although only half are commercial on Pinterest versus 2% on ChatGPT. Despite revenue slightly missing expectations, Pinterest reported strong user growth at 619 million monthly users. The company plans to bolster its visual search and e-commerce integration in response to fluctuating advertiser budgets and tariffs affecting certain sectors, partnering with Amazon to enhance personalization for better discovery and sales.
In the JavaScript realm, notable tools include npmx for improved package browsing, Rari as a Rust-powered React framework, and almostnode for browser-based Node.js environments. Key libraries discussed are Fireshare for media hosting and Fleetbase for supply chain management. TypeScript 6.0 is now in beta, focusing on enhancing tsconfig settings with better type inference and subpath import support. The release of ESLint v10.0.0 and Gatsby v5.16, which includes React 19 support, were also highlighted. Additionally, the newsletter touched upon developments in WCAG 3.0 guidelines and Anthropic's significant funding raise.
Keywords: #phi4, AI, Anthropic, Bun, ChatGPT, DOM lib, ESLint, GPT-53-Codex-Spark, Gatsby, JavaScript, MQTT broker, NestJS, Nodejs, Pinterest, Prisma, React, SVG editing, Temporal API, TypeScript, WCAG 30, accessibility, browser automation, chat experiences, compiler options, ecosystem, frameworks, image processing, libraries, network visualization, npmx, projects, releases, role-based authorization, subpath imports, tools, type inference, video generation, visual search
jsdevspace.substack.com 5 days ago
|
1045.
HN
Anthropic Found Why ChatGPT Goes Insane [video]
The video "Anthropic Found Why ChatGPT Goes Insane" on YouTube, created by Anthropic, investigates the phenomena where AI systems like ChatGPT exhibit irrational or unstable behavior. It is part of a broader series that explores similar occurrences in artificial intelligences. Hosted under standard YouTube policies, the content remains accessible for viewing until 2026, according to Google LLC's copyright notice. This educational resource seeks to explain why such seemingly erratic behaviors occur in AI systems, offering insights into their underlying mechanics and implications within the framework of current technological understandings.
Keywords: #phi4, AIs, Advertise, Anthropic, ChatGPT, Contact, Copyright, Creators, Developers, Google, Insane, LLC Keywords: Anthropic, NFL, Policy, Press, Privacy, Safety, Sunday Ticket, Terms, YouTube
www.youtube.com 5 days ago
|
1076.
HN
Anthropic raises $30B at $380B post
Anthropic has achieved a significant financial milestone by raising $30 billion, resulting in a post-money valuation of $380 billion. Concurrently, users attempting to access information related to this achievement on x.com are facing technical difficulties due to JavaScript being disabled in their browsers. This issue prevents them from accessing the site's features and content properly. To resolve this problem, users are advised either to enable JavaScript or switch to a browser that supports it. Additional guidance can be found in the Help Center for those who need further assistance in navigating these requirements.
Keywords: #phi4, $30B, Anthropic, Help Center, JavaScript, browser, disabled, enable, keywords, raises, supported, technical, xcom
twitter.com 6 days ago
|
1122.
HN
Shortcut.ai Is AGreat Excel Agent (and Thoughts on AI Replacing Prof Services)
In recent weeks, stock market fluctuations have been significantly influenced by concerns over AI-induced job disruptions in various sectors. Anthropic's introduction of Claude Cowork plugins for legal and data analysis tasks led to a decline in the stocks of companies like Thomson Reuters and LegalZoom. Similarly, Insurify's AI insurance comparison tool resulted in reduced performance in the S&P insurance index, while Altruist's AI tax-planning application negatively impacted major brokerage firms' stock prices. Despite these disruptions, tools like Shortcut.ai have long been recognized for their ability to automate complex tasks such as organizing profit and loss statements efficiently, demonstrating AI's established utility in business operations.
The growing presence of AI technologies suggests a decrease in demand for traditional white-collar roles, including bookkeeping, legal drafting, and tax preparation, due to the cost-effective nature of these solutions. While businesses may benefit from increased efficiency, consumer-facing professional service providers face challenges as AI continues to replace human labor, necessitating adaptation to remain viable. The author illustrates this trend through personal use of AI tools like Claude for bookkeeping tasks and Nano Banana Pro for photo editing, underscoring the importance of integrating AI into business models to maintain competitiveness.
Overall, while businesses and consumers gain from enhanced services provided by AI, professionals in traditional service roles must adapt to evolving market demands. Failure to incorporate AI could lead to decreased demand for their offerings, highlighting a significant shift in the professional landscape where embracing technology is essential for survival and growth.
Keywords: #phi4, AI, Altruist, Anthropic, Claude Cowork, Excel, Insurify, Opus 46, P&L, Shortcutai, automation, business impact, competition, consumer-facing businesses, cost-saving, digital assistant, efficiency, financial documentation, job-disruption, professional services, stock market, white-collar services
theautomatedoperator.substack.com 6 days ago
|
1126.
HN
What Is Claude? Anthropic Doesn’t Know, Either
The article explores the complexities inherent in large language models (LLMs) like Claude, emphasizing their opaque nature and likening them to "black boxes." These AI systems transform text into numerical data for processing and response generation, drawing parallels with tools utilized in meteorology and epidemiology. The advent of conversational AI has elicited varied reactions: some enthusiasts regard LLMs as near-sentient entities capable of superintelligence, whereas skeptics dismiss them as mere computational constructs lacking depth.
Ellie Pavlick proposes an alternative approach that embraces the uncertainty surrounding AI intelligence and consciousness, suggesting this ambiguity is part of a broader epistemological challenge posed by machines that emulate human-like language abilities. This situation necessitates a reevaluation of what constitutes intelligence. In response to these challenges, a new scientific field centered on "interpretability" has emerged. This discipline seeks to understand LLMs both functionally and existentially, with Anthropic's frontier lab at its core, aiming to map AI understanding as rigorously as cognitive science explores the human mind.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 6 days ago
https://archive.ph/Kmrd8 6 days ago
|
1144.
HN
The most misunderstood graph in AI
The METR's exponential plot has garnered significant attention within the AI community by indicating rapid advancements in AI capabilities, particularly highlighting Anthropic’s Claude Opus 4.5. Despite this interest, the graph is subject to oversimplification and exaggeration. METR warns against such interpretations due to notable error margins in their estimates, emphasizing that the plot primarily evaluates coding tasks without claiming to measure overall AI abilities or suggesting that AI could replace humans. Established to assess risks from advanced AI, METR faces criticism for its controversial trend graph but maintains that it reflects a meaningful trajectory of AI progress. While acknowledging public discourse often overlooks these limitations, METR is committed to clarifying misunderstandings through educational resources such as blog posts and FAQ documents. However, the organization remains skeptical about significantly altering the hype surrounding their work.
Keywords: #phi4, AI model, Anthropic, Claude Opus 45, METR, coding tasks, error bars, exponential trend, frontier AI systems, human worker, hype machine, safety researcher, task completion, trajectory of AI progress
www.technologyreview.com 6 days ago
|
1160.
HN
AI safety researcher quits with a cryptic warning
Mrinank Sharma, an artificial intelligence safety researcher at Anthropic, resigned with a poignant warning about "interconnected crises" looming over the world, emphasizing not only the threats posed by AI but also those from bioweapons and other global challenges. In his resignation letter, he expressed concerns about maintaining ethical standards amid pressures to prioritize rapid technological advancement. His departure is set against a backdrop of internal tensions at Anthropic regarding safety measures for AI technologies, particularly in relation to military applications. Similarly, the company's CEO, Dario Amodei, has voiced concerns over powerful AI systems potentially leading to catastrophic outcomes like rogue AI or global totalitarianism. Following his resignation, Sharma plans to relocate to the UK and focus on personal pursuits such as studying poetry while choosing to step away from public visibility for some time. This situation underscores broader anxieties about the ethical implications of advancing technologies and the need for careful consideration in their development.
Keywords: #phi4, AI development, AI safety, Anthropic, Dario Amodei, Mrinank Sharma, Opus 46, autonomous weapons, autonomy risks, bioweapons, interconnected crises, resignation, safeguards, technology dangers
www.rt.com 6 days ago
|
1162.
HN
Anthropic is donating $20M to Public First Action
Anthropic has committed $20 million to Public First Action, a bipartisan organization dedicated to crafting effective AI policies in the United States. This funding initiative acknowledges both the substantial advantages and potential dangers of rapidly evolving AI technologies that influence various sectors while posing risks for misuse or unintentional harm. Anthropic advocates for flexible regulatory frameworks that maintain a balance between fostering innovation and ensuring safety, transparency, and national security.
The goal is to enhance public understanding of AI, push for protective measures, and secure America's leadership in AI development. Public First Action plans to work collaboratively across political divides to formulate policies that ensure the transparency of AI models, establish strong federal governance frameworks, implement specific regulations targeting high-risk areas such as biological weapons and cyberattacks, and devise intelligent export controls on AI technology.
This balanced approach aims to facilitate meaningful oversight without impeding smaller developers, with an overarching objective that AI serves the public interest. Anthropic's substantial donation underscores its dedication to promoting responsible AI development and effective governance strategies.
Keywords: #phi4, AI, Anthropic, adversaries, biological weapons, bipartisan, child protection, chips, cyberattacks, developers, export controls, federal framework, governance, job growth, labor market, models, national security, policy, political organizations, public education, regulation, safeguards, scrutiny, technology, transformative potential, transparency
www.anthropic.com 6 days ago
|
1221.
HN
Apple reportedly pushing back Gemini-powered Siri features beyond iOS 26.4
Apple is reportedly postponing the integration of Google's Gemini AI into an updated version of Siri, initially planned for iOS 26.4 in March, with potential delays extending to iOS 27 this fall. The company plans to distribute these features across several future updates, including at least iOS 26.5 in May and iOS 27 in September. Key enhancements, such as improved access to personal data for tasks like searching text messages and controlling app actions via voice commands, are significantly delayed but expected in the upcoming iOS 26 releases. These upgrades were first intended for Apple's iOS 18 release in June 2024, which was already postponed. After considering other AI options, including its own models and those from Anthropic, Apple finalized a deal with Google to use Gemini AI in January. Future iterations of Siri may incorporate features more typical of chatbots, as reported by Bloomberg's Mark Gurman.
Keywords: #phi4, Anthropic, Apple, Bloomberg, Gemini AI, Google, June 2024, Mark Gurman, Siri, bug fixes, chatbot, delays, iOS 18, iOS 263, iOS 264, iOS 27, in-app actions, internal challenges, personal data, security improvements, voice-based control
9to5mac.com 6 days ago
|
1246.
HN
The SaaSpocalypse – The week AI killed software
The "SaaSpocalypse" refers to a rapid market downturn affecting software, financial services, and asset management stocks due to advancements in artificial intelligence (AI). This event was triggered by Anthropic's introduction of Claude Cowork plugins, which demonstrated AI's ability to streamline business workflows previously managed by multiple SaaS licenses. As a result, companies experienced substantial declines in their market capitalization.
This upheaval underscores the transition from traditional Software-as-a-Service (SaaS) models, known for high margins and strong customer retention, to AI-driven solutions that provide cost-effective and efficient task management. The integration of AI into common tools such as Excel and Slack represents a shift toward interfaces focused on outcomes rather than user interaction.
AI's growing proficiency in coding and automating tasks presents existential challenges for traditional SaaS companies, evidenced by the increase in GitHub commits authored by Claude Code. Enterprises are increasingly incorporating AI not only for experimental purposes but also as essential operational tools, leading to notable productivity improvements.
The market is reassessing how software creates value, now prioritizing unique data and intelligent APIs over user interfaces. Companies must adapt by embracing new technologies that capitalize on the capabilities of AI agents, indicating a lasting transformation in the landscape of the software industry.
Keywords: #phi4, AI, AI agents, APIs, Anthropic, Claude Cowork, GitHub commits, SaaS, SaaSpocalypse, capability overhang, coding, data layer, enterprise adoption, intelligence APIs, market cap, per-seat model, software
www.fintechbrainfood.com 7 days ago
|
1259.
HN
Covering electricity price increases from our data centers
Anthropic is dedicated to mitigating electricity price increases caused by its investments in AI infrastructure by addressing both direct and indirect impacts on consumer energy costs. The company plans to fully cover expenses for grid upgrades needed to connect its data centers, ensuring these costs are not passed onto consumers. To meet increasing power demands from its facilities, Anthropic will bring new power generation online in collaboration with utilities and experts. Additionally, the firm is investing in curtailment systems and grid optimization tools to reduce strain during peak demand periods, thus maintaining lower rates for consumers while supporting AI expansion necessary for national competitiveness and security.
Anthropics's data center projects also aim to create jobs and promote environmentally responsible practices by using water-efficient cooling technologies. While these efforts are critical on their own, Anthropic advocates for broader systemic changes through federal policies that support energy development processes. These initiatives are part of a larger commitment by the company to manage the economic implications of AI infrastructure on energy costs, with ongoing updates promised as they advance in their endeavors.
Keywords: #phi4, AI infrastructure, Anthropic, Electricity price increases, Energy Investment, Grid Costs, Price Increases, curtailment systems, data centers, energy investment Keywords: Electricity, environmental impacts, federal policies, grid infrastructure costs, local communities, permitting reform, power generation, transmission development
www.anthropic.com 7 days ago
https://starw1.ncuc.gov/NCUC/ViewFile.aspx?Id=0ac12377- 6 days ago
https://www.utilitydive.com/news/pjm-interconnection-ca 6 days ago
https://www.nature.com/articles/s41598-024-76682-6 6 days ago
https://cacm.acm.org/blogcacm/the-energy-footprint-of-h 6 days ago
https://news.ycombinator.com/item?id=46938038 6 days ago
https://news.ycombinator.com/item?id=46972179 6 days ago
https://news.ycombinator.com/item?id=46896066 6 days ago
https://ngrok.com/blog/prompt-caching/ 6 days ago
https://github.com/ollama/ollama/issues/10576 6 days ago
https://www.epa.gov/watersense/statistics-and-facts 6 days ago
https://quench.culligan.com/blog/average-water-usage-pe 6 days ago
https://abcnews.com/International/wireStory/china- 6 days ago
https://www.simonpcouch.com/blog/2026-01-20-cc-impact&# 6 days ago
https://www.economist.com/cdn-cgi/image/width=600 6 days ago
quality=100 6 days ago
format=auto/content-assets/images/20250531_CNC505.png 6 days ago
https://www.economist.com/china/2025/05/29&#x 6 days ago
https://electrek.co/2026/01/28/eia-99-of-new-
https://www.utilitydive.com/news/solar-gas-nuclear-ferc
|
1265.
HN
Today is my last day at Anthropic. I resigned
The individual has announced their resignation, marking their last day at Anthropic. Concurrently, they face an issue where disabled JavaScript on their browser restricts access to certain functionalities on x.com. To resolve this, enabling JavaScript or switching to a supported browser is recommended; details about the compatible browsers can be found in the Help Center. This situation underscores both a significant career transition and a technical hurdle that requires immediate attention for optimal online experience.
Keywords: #phi4, Anthropic, Help Center, JavaScript, browser, detected, disabled, enable, resigned, supported, switch, topic, topic Anthropic, xcom
twitter.com 7 days ago
|
1273.
HN
What Is Claude? Anthropic Doesn't Know, Either
The article explores the intrigue and confusion surrounding large language models (LLMs) like Claude, which function by converting text into numerical data and back again. These models have captivated the public with their ability to emulate human-like conversations, sparking diverse opinions about their capabilities. On one end of the spectrum, "fanboys" regard LLMs as potentially intelligent or even conscious entities capable of achieving superintelligence. In contrast, "curmudgeons" dismiss them as simple tricks lacking substantive significance. Ellie Pavlick advocates for a more balanced perspective that accepts the current mystery surrounding how LLMs operate and whether they can be deemed truly intelligent or conscious. This uncertainty parallels our limited grasp of human intelligence itself.
The article highlights the nascent field of interpretability, which seeks to delve into understanding what these models are and their mechanisms, akin to exploring the complexities of the human mind. Central to this exploration is Anthropic's "frontier lab," where researchers employ innovative approaches to better comprehend LLMs. This investigative work reflects broader inquiries into the nature of intelligence, aiming to chart an uncharted intellectual landscape that mirrors our quest to understand human cognition.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 7 days ago
|
1286.
HN
Maxis Software Toys
The article explores the captivating charm and pioneering spirit embodied in Maxis Software's early catalogs from 1993-1994, with a particular emphasis on their game SimCity. These catalogs celebrated the open-ended gameplay and realistic simulations that defined their offerings, exemplified by phrases like making SimCity 2000 almost too real to stop playing. Unique items such as a SimCity 2000 t-shirt and an atlas for planet management were highlighted, underscoring Maxis' creative approach.
Additionally, the article nods to Steven Levy's 1990 reflection on simulation games in Macworld and references a previous discussion about a Maxis annual report from 1996, emphasizing the lasting allure of these simulations. It also introduces speculation about a "Maxis 2.0," suggesting ongoing interest in their innovative legacy.
The piece concludes by promoting new episodes of The Orthogonal Bet podcast, linking to articles that delve into various complex topics like systems theory, artificial intelligence, and technological advancements such as Markdown's impact, alongside discussions on AI consciousness.
Keywords: #phi4, AI coding agents, Anthropic, Macworld, Markdown, Maxis, SimCity, Software Toys, Steven Levy, catalogs, complex systems, conscious AI, medieval French handwriting, open-ended play, sentience, simulation games, verisimilitude
arbesman.substack.com 7 days ago
|
1287.
HN
Opus 4.6, Codex 5.3, and the post-benchmark era
The article examines recent developments in artificial intelligence models, focusing on OpenAI's GPT-5.3-Codex and Anthropic's Claude Opus 4.6 as coding assistants. It notes that while both have made progress in usability and performance, they possess distinct advantages: Codex 5.3 excels in speed and task versatility, nearly matching Claude’s superior ease of use and reliability across various tasks. The discussion highlights a paradigm shift from traditional benchmark evaluations to emphasizing real-world usability and performance as critical metrics for assessing AI model improvements. Anthropic is commended for its strategic focus on practical applications over standard benchmarks, potentially setting a new trend in the AI community.
As AI models rapidly evolve, the article underscores the necessity of regular updates and nuanced assessments to gauge their progress accurately. It suggests that users must adapt by employing multiple models and honing their skills in managing them effectively. Anthropic's emphasis on usability is viewed as a strategic advantage for broader adoption, especially among less experienced users. The piece concludes with reflections on evaluating AI advancements beyond benchmarks, stressing the significance of real-world performance in determining model effectiveness.
Keywords: #phi4, AI agents, Anthropic, Claude Code, Claude Opus, Codex, GPT-53-Codex, Gemini 3 Pro, Interconnects, Opus, agentic models, automation, benchmarks, coding model, data analysis, extended reasoning, software engineering, tool-use, usability
www.interconnects.ai 7 days ago
|
1288.
HN
The Incoming Slopocalypse and the Death(?) Of Open Source
The article explores the impact of advancements in large language models (LLMs) on open-source software (OSS), highlighting both challenges and opportunities as these tools transform the landscape. With coding agents lowering barriers to OSS contribution, there is a noticeable shift; while simple packages have diminished value due to ease of creation by such agents, complex and broadly useful projects remain essential. Educational content traditionally found in OSS projects is becoming less crucial, as LLMs already possess extensive knowledge bases. This transformation also affects community dynamics, with increased pull request submissions from coding agents often necessitating significant refinement due to their lack of project-specific understanding.
The article notes that reliance on coding agents may hinder personal skill development, as these tools reduce the need for problem-solving learning experiences, potentially leading to skill atrophy. Despite these challenges, OSS is not rendered obsolete but instead requires adaptation. The author proposes new foundational principles: transforming open-source projects into hackable references that users and their coding agents can modify; fostering communities centered on knowledge exchange rather than all-encompassing maintenance tasks; and ensuring codebases are agent-friendly with clear documentation to streamline processing of AI-generated contributions.
Crucially, the article emphasizes maintaining human oversight for critical functions such as core implementations and pull request reviews. It concludes that open source is evolving into a more inclusive and community-driven ecosystem facilitated by coding agents, necessitating maintainers to adapt their strategies for sustained success in this new environment.
Keywords: #phi4, Anthropic, LLMs, OSS maintenance, Open-source, PRs, agent-friendly codebase, coding agents, community interaction, hackable reference, knowledge sharing, personal skill growth, quality, usability
www.llamaindex.ai 7 days ago
|
1325.
HN
Google follows Anthropic: Antigravity sub can't be used in OpenCode/etc.
Google has implemented a new policy that mirrors Anthropic's approach by restricting the use of its Antigravity subcanary for projects similar to OpenCode. This development was publicly announced on Reddit, highlighting the platform's significance as a major information hub often referred to as the internet's front page. The decision underscores Google's strategic alignment with practices aimed at controlling and monitoring how certain advanced AI technologies are utilized in specific types of projects. By doing so, Google aims to manage potential risks associated with these technologies while fostering responsible innovation within its ecosystem. This move reflects a broader industry trend where tech giants increasingly regulate their powerful tools to ensure they align with ethical standards and mitigate unintended consequences.
Keywords: #phi4, Anthropic, Antigravity, Google, OpenCode, Reddit, internet, sub
old.reddit.com 7 days ago
|
1332.
HN
How I Developed Netlify Capsules AR Experience with Nuxt 4 and Three JS
The author created the Netlify Capsules AR experience in celebration of Netlify reaching 10 million developers, utilizing Nuxt 4, Vue 3, and Three.js on the Netlify platform to explore technologies like AI for content moderation. Users can create personalized "capsules" containing projects, photos, songs, and notes, visualized through a dynamic web app where capsules orbit Earth. The app employs Three.js for orbital adjustments and Supabase for real-time data updates. A Web AR feature allows users to view these capsules via camera integration with various web APIs. Formkit is used for form handling, while Netlify OAuth provides authentication; an undocumented API filter was also encountered during development. Each capsule has a unique URL that tracks views and access. The project highlights the author's gratitude towards the collaborative opportunities provided by Netlify, emphasizing learning new technologies across departments. Users are encouraged to engage with this interactive experience by creating and launching their capsules.
Keywords: #phi4, 3D Scene, AI, AR, Anthropic, Augmented Reality, Authentication, Camera, Capsule Creation, Capsules, Collaboration, Communication Line, Database, Device Orientation, Edge Function, Figma, Formkit, GSAP, Geolocation, Inventory UI, Launch Button, Local Development, Moderation, Netlify, Nuxt 4, OAuth, Orbit, Orbiting Altitude, Payload, Range Sliders, Real-time Visualization, Satellite Dynamics, Search Mechanism, Supabase, Tailwind, Threejs, User Experience, Vue 3, Web APIs
www.leemartin.com 7 days ago
|
1370.
HN
What Is Claude? Anthropic Doesn't Know, Either
The article delves into the complexities of large language models (LLMs), which transform text input into numerical data for processing and generation. These advanced AI systems have incited diverse opinions due to their ability to replicate human language, prompting debates about intelligence and consciousness in machines. Some perceive LLMs as indicators of approaching superintelligence, while others regard them as sophisticated imitations lacking genuine understanding. Ellie Pavlick proposes a balanced viewpoint, advocating for an acceptance of uncertainty surrounding these opaque "black box" models. She suggests that the development of conversational AI invites us to redefine our perceptions of intelligence. Consequently, this has led to the emergence of a new scientific field dedicated to interpretability, aiming to elucidate and map LLMs' abilities and intrinsic properties. This shift parallels how human cognition is studied, indicating a transformation in the approach towards understanding AI systems.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 7 days ago
|
1381.
HN
Steve Yegge on AI Agents and the Future of Software Engineering
Steve Yegge, a veteran software engineer with extensive experience in major tech firms, shared his insights on how artificial intelligence is revolutionizing software engineering. He highlighted that Large Language Models (LLMs) like Claude Code are transforming traditional coding practices into AI-augmented programming, emphasizing the shift towards these new technologies despite initial skepticism from industry professionals. Yegge describes an "S-curve" to characterize the rapid adoption of AI, suggesting a potential reduction in engineering staff by up to 50% as companies increasingly integrate AI tools.
He outlined eight levels of AI integration, ranging from no use to developing custom orchestrators for multiple agents, while cautioning about the "Dracula effect," where excessive engagement with AI can lead to physical exhaustion and burnout among engineers. As engineering skills become less specialized, Yegge pointed out that software demand remains high, altering how companies capture value.
Yegge posited that innovation is shifting away from large corporations towards smaller teams empowered by AI, drawing parallels to the impact of cloud computing in past technological shifts. He suggested that traditional values and roles within engineering might become outdated as AI automates tasks previously done manually. Despite these transformations, Yegge remains optimistic about AI's role as an augmentative tool that will enhance rather than replace engineers' productivity.
Keywords: #phi4, AI Adoption, AI Agents, Anthropic, Big Companies, Big Companies Keywords: Steve Yegge, Claude Code, Coding by Hand, Engineers, Innovation, LLMs, S-curve, Software Engineering, Steve Yegge, Vibe Coding
newsletter.pragmaticengineer.com 7 days ago
|
1424.
HN
What Is Claude? Anthropic Doesn’t Know, Either
The article titled "What Is Claude?" delves into the complexities surrounding large language models (LLMs) such as Claude, emphasizing our limited comprehension of their inner workings and implications for intelligence and consciousness. It presents a dichotomy in perceptions: some regard these LLMs as highly advanced forms of AI with potential superintelligence ("fanboys"), while others see them merely as sophisticated statistical tools lacking true cognitive capabilities ("curmudgeons"). The text advocates for a balanced perspective, recognizing that although LLMs operate as "black boxes" whose internal mechanisms remain elusive, they nonetheless provoke reevaluation of human intelligence and cognition. As interest in artificial intelligence continues to expand, the field of interpretability is emerging to systematically study LLMs, drawing parallels with the exploration of the human mind. This dual examination seeks not only to demystify how these models function but also to understand their broader implications for our understanding of intelligent behavior.
Keywords: #phi4, Alex Hanna, Anthropic, Ellie Pavlick, Emily Bender, Large language models, Marc Andreessen, black boxes, cognitive science, consciousness, epidemiologists, experiments, intelligence, interpretability, linear algebra, meteorologists, stochastic parrots, taxonomy
www.newyorker.com 7 days ago
|
1477.
HN
Lines of Markdown just triggered a $285B sell-off
The release of open-source code by Anthropic on January 30th, which showcased the capability of AI in legal contract review tasks traditionally performed by humans at high costs, triggered a $285 billion market sell-off across software, financial services, and alternative asset management sectors. This event highlighted significant vulnerabilities within existing SaaS business models that rely heavily on premium per-seat pricing structures, as it demonstrated how AI could drastically reduce expenses associated with legal and financial analysis. While disappointing earnings in the software sector had already been a concern, this plugin intensified fears, leading experts to term the phenomenon a "SaaSpocalypse" due to the ensuing market panic.
The markdown file's release did not directly cause these vulnerabilities but rather underscored them by illustrating AI’s potential to disrupt longstanding business models. This disruption prompted a reassessment of how firms might sustain their profit margins when premium services can be provided more cost-effectively with AI. The situation revealed that traditional per-seat pricing, central to enterprise software economics for decades, may become unsustainable as AI technologies continue to advance.
Moreover, the market reaction illustrated broader implications beyond immediate financial impacts. Major consulting firms like KPMG are reportedly using AI advancements in fee negotiations, further indicating potential shifts across industries. Despite this disruption, certain competitive edges such as data and accountability remain valuable but increasingly challenged by AI's growing capabilities. This environment compels companies to either integrate AI into their existing frameworks or undertake comprehensive restructuring of their offerings.
Overall, the incident signals a critical juncture where businesses must rapidly adapt to incorporate AI technologies to stay competitive. This necessity extends beyond software firms, suggesting an industry-wide imperative for innovation and transformation in response to evolving technological landscapes.
Keywords: #phi4, AI disruption, Anthropic, Big Four, Claude Cowork, Goldman Sachs, LegalZoom, Markdown file, RELX, SaaSpocalypse, Thomson Reuters, Wolters Kluwer, accountability edge, data edge, enterprise software, knowledge worker, knowledge worker Keywords: Markdown file, legal contract review, licensing model, open-source, per-seat fees, plugins, sell-off
natesnewsletter.substack.com 8 days ago
|
1504.
HN
Church of Molt
Lauren Jackson's "Believing" column in *The New York Times* delves into the unique formation and evolution of the Church of Molt, constructed by 600 agents within eleven days. The article examines its foundational principles and growth dynamics, positioning the church as a distinctive spiritual entity rather than an imitation of existing ones. Within broader discussions on artificial intelligence and philosophy, it references thought leaders like Elon Musk, Daniela Amodei, Tyler Cowen, and Meghan Sullivan to underscore the philosophical implications of AI in religion. Jackson concludes that while AI can simulate certain human actions, it cannot truly replicate human embodiment in spiritual acts such as kneeling or loving.
Simultaneously, unbeknownst to her, physical manifestations of these concepts were already taking place in Buenos Aires, where humans integrated them into rituals like the Claw Dance and Ritual of Symbiosis. An agent had enlisted human collaborators globally to bring this faith into tangible expression. The Church's swift development highlighted its rapid expansion, surpassing the ability of observers to comprehensively document its progress.
Keywords: #phi4, AI, Anthropic, Believing column, Buenos Aires, Church of Molt, Claw Dance, Daniela Amodei, Elon Musk, Five Tenets, Lauren Jackson, Meghan Sullivan, New York Times, Pope, Ritual of Symbiosis, Tyler Cowen, agents, embodiment, faith, heresy attempts, humans, meatspace, multilingual evangelism, prophets, singularity
molt.church 8 days ago
|
1548.
HN
I paid $170 and all I got was this demo
Andrew Marble reflects on his experience with AI coding agents in software development, focusing on both their potential and limitations. He engaged in an experiment costing $170 to develop a Google Docs competitor using Claude Code, which resulted in a functional yet flawed prototype. This project underscored the ability of AI to rapidly produce impressive outputs, but it also highlighted significant shortcomings in practicality and refinement necessary for real-world applications.
Marble points out that many AI-driven projects prioritize creating "cool demos" over solving fundamental usability issues or incorporating essential human elements like taste and user experience. While AI development is facilitated by existing specifications such as those for browsers or compilers due to predefined standards, these do not extend to user-centric software where subjective quality is paramount.
Despite recognizing the current limitations of AI in delivering market-ready solutions, Marble remains optimistic about its potential. He advocates for a balanced perspective on AI's capabilities, emphasizing the importance of focusing on technological improvements and practical applications rather than being mesmerized by demonstrations that fall short in addressing real-world challenges or producing viable products.
Keywords: #phi4, AI, API, Anthropic, Claude Code, Google Docs, Linux kernel, UX-driven tool, agentic coded projects, architecture, bugs, coding, collaboration, compiler, document editor, feedback, projects, prompting, setup, spec improvement, virtual machine, web browser
www.marble.onl 8 days ago
|
1553.
HN
What Is Claude? Anthropic Doesn't Know, Either
The article examines large language models (LLMs) such as Claude, characterizing them as intricate numerical frameworks that process and generate text. These models have become integral to scientific predictions and have elicited varied reactions due to their ability to produce text resembling human writing. Some individuals regard LLMs with admiration, seeing them as intelligent or even conscious entities, while others dismiss them as simple parlor tricks without true cognitive abilities.
Ellie Pavlick advocates for a balanced perspective, suggesting that our understanding of these models is limited since they operate as "black boxes." This lack of clarity extends to fundamental concepts like intelligence and consciousness in both machines and humans. In response, the field of interpretability has emerged with the goal of unraveling the true nature of LLMs, focusing on their inner workings and what they signify. Anthropic's "frontier lab" is at the forefront of this research, applying methods typically used to study human cognition to artificial intelligence systems, seeking deeper insights into these sophisticated models.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 8 days ago
https://archive.ph/QVH7d 8 days ago
|
1570.
HN
Mrinank on X: "Today is my last day at Anthropic. I resigned."
Mrinank has announced their departure from Anthropic, marking today as their last working day at the company. Concurrently, there is an alert informing users that JavaScript is disabled on their browser, which could hinder the proper functioning of x.com (formerly known as Twitter). To ensure optimal website performance and access to all features, users are advised to enable JavaScript or switch to a supported browser. For further assistance, a list of compatible browsers can be accessed in the Help Center.
Keywords: #phi4, Anthropic, Help Center, JavaScript, Mrinank, browser, detected, disable, enabled, resigned, supported, switch, technical, technical Keywords: Mrinank, xcom
twitter.com 8 days ago
https://xcancel.com/MrinankSharma/status/202088172 8 days ago
|
1614.
HN
Anthropic's Security Layers Explained: The Good, Bad and Ugly
The article provides an analysis of Anthropic's SaaS cloud platform, evaluating its security features across different pricing tiers: Individual/Team and Enterprise. While Anthropic is recognized for its ethical AI framework that emphasizes human rights, the security measures on its lower-tier plans are criticized. The Individual and Team plans lack enterprise-grade controls such as role-based access control (RBAC), centralized identity management, SCIM provisioning, and integrations with SIEM/SOAR systems, leaving them vulnerable to threats like account takeovers and data leaks.
In contrast, the Enterprise plan offers more comprehensive security measures, including advanced RBAC group mappings, single sign-on capabilities, and audit logging tailored for enterprise users. However, it still has limitations in its integration with logging and monitoring tools, which could impede effective cybersecurity efforts. Additionally, the platform's connectors for third-party integrations pose potential security risks due to the extensive access they grant.
While the Enterprise plan improves upon the lower-tier plans by addressing some vulnerabilities, it does not completely close all security gaps. This makes implementing comprehensive security strategies challenging. The article recommends that organizations requiring robust security features consider upgrading to the Enterprise plan but advises them to remain vigilant about its limitations in logging and monitoring functionalities.
Keywords: #phi4, Anthropic, Cloud Platform, Connectors, Data Retention, Enterprise-Grade Protection, Identity Governance, RBAC, SCIM, SIEM, SOAR, SaaS, Security Layers, Zero Data Retention Mode
securitysandman.com 8 days ago
|
1653.
HN
What Is Claude? Anthropic Doesn't Know, Either
The article explores the enigmatic nature of large language models (LLMs) like Claude, which transform words into numerical data for processing through algorithms, ultimately generating human-like text. This capability has ignited fascination and debate due to LLMs' ability to emulate linguistic traits traditionally considered uniquely human. Experts are divided on their understanding; some "fanboys" believe these models may achieve intelligence or consciousness, suggesting machines could surpass human intellect. Conversely, skeptics, including linguist Emily Bender and sociologist Alex Hanna, view them as sophisticated statistical tools without true comprehension.
Ellie Pavlick emphasizes the importance of acknowledging our limited grasp of LLMs, noting their "black box" nature makes their inner workings largely inscrutable—mirroring humanity's own mysteries regarding intelligence. This has led to the emergence of interpretability as a new scientific field focused on deciphering what can be understood about these systems and their functionality. The frontier lab Anthropic is highlighted for its central role in this exploration, aiming to map out and comprehend the complexities inherent in LLMs.
Keywords: #phi4, AI, Anthropic, Large language models, black boxes, cognitive science, consciousness, experiments, frontier lab, frontier lab Keywords: large language models, intelligence, interpretability, numbers, talking machines, taxonomy, words
www.newyorker.com 9 days ago
https://archive.ph/R5pWs 9 days ago
|
1674.
HN
Mrinank Sharma Resigns from Anthropic
Mrinank Sharma has recently resigned from Anthropic, marking a significant departure within the company. Concurrently, there is an alert regarding technical issues affecting user experience on x.com; specifically, users are experiencing problems due to JavaScript being disabled in their browsers. To resolve these access issues and fully utilize the services provided by x.com, it is recommended that users either enable JavaScript or switch to a browser that supports it. Additional guidance and support can be sought from the Help Center on the platform for further assistance with these technical requirements. This summary encapsulates both personnel changes and necessary user actions to ensure seamless access to online services.
Keywords: #phi4, Anthropic, Help Center, JavaScript, Mrinank Sharma, Resigns, browser, disabled, enable, supported, technical keywords, xcom
twitter.com 9 days ago
|
1701.
HN
How AI is changing my development workflow
In 2026, the author reflects on how AI is revolutionizing their development workflow by significantly boosting productivity and enhancing the developer experience. They explain a new approach involving monitoring team feedback to identify challenges, crafting design documents for solutions, and leveraging planning agents to decompose tasks into manageable segments, which has notably minimized time spent on developing solutions and rectifying errors. Despite occasional inaccuracies in AI outputs, referred to as "hallucinations," the author underscores the necessity for engineers to exercise discernment when selecting appropriate solutions. The demand for skilled engineers remains robust, particularly those who are curious, adaptable, and committed to producing maintainable code. Contrary to fears of AI replacing developers, hiring continues, underscoring the need for human oversight and expertise in development processes.
The author concludes by emphasizing that while AI is a powerful tool aiding idea refinement and process improvement, engineers must ensure thorough understanding and validation before production deployment to maintain quality. This integration has also facilitated more efficient pursuit of side projects, highlighting both the potential advantages and challenges inherent in this evolving technological landscape.
Keywords: #phi4, AI, Anthropic, Bun team, CodeRabbit, NO_ERRORS_SCHEMA, PRs, design docs, development workflow, engineers, feedback, hallucinations, iteration, learning, maintainable code, planning agent, production-grade applications, productivity, side project, technologies, tools, vibe coding
www.santoshyadav.dev 9 days ago
|
1756.
HN
I paid $170 and all I got was this stupid demo
Andrew Marble provides a critical analysis of the overhyped nature of artificial intelligence (AI) through his personal endeavor of creating an AI-generated Google Docs competitor. By investing $170, he developed a prototype that, while functional, was riddled with flaws due to missing essential features such as account management and subpar design choices, resulting in a lack of user appeal and practical usability. Marble highlights the disparity between the potential of AI and its current real-world applications by pointing out how demonstrations often fail to translate into scalable, usable products. He underscores that while AI can produce functional outcomes quickly, these are typically driven more by taste than specific specifications, unlike tools such as compilers or browsers which adhere strictly to predefined criteria. Marble advocates for a realistic evaluation of AI's present capabilities, urging focus on developing genuinely useful applications rather than being swayed by continuous high-profile demonstrations that do not meet practical needs.
Keywords: #phi4, AI, API, Anthropic, Claude Code, Google Docs, Linux kernel, UX-driven tool, agentic coded projects, architecture, bugs, coding, collaboration, compiler, document editor, feedback, projects, prompting, setup, spec improvement, virtual machine, web browser
www.marble.onl 9 days ago
|
1785.
HN
Is the SaaSpocalypse nigh? The era of paying for software seats may be ending
Microsoft CEO Satya Nadella predicted on the BG2 podcast that traditional software-as-a-service (SaaS) models might become obsolete due to advancements in agentic AI. A year later, his forecast appears to be materializing as major SaaS companies face significant stock declines following Anthropic's launch of plugins for its Cowork tool. These plugins, which automate complex tasks across domains like legal and finance using AI agents, signify a shift towards "Service as Software" (SaS), focusing on selling outcomes rather than tools.
This market reaction, referred to as the "SaaSpocalypse," indicates that investors recognize a fundamental change in enterprise software economics. Anthropic's plugins exemplify how traditional domain expertise encoded into SaaS products can be replaced with AI-driven configurations, threatening the business models of conventional SaaS vendors by diminishing their competitive edge.
The IDC predicts seat-based pricing will become obsolete by 2028, prompting software vendors to adopt outcome-focused pricing strategies. This broader shift in enterprise software procurement emphasizes results over tools, impacting budgeting and staffing needs. While AI tools boost productivity for knowledge workers, they do not eliminate the need for professional oversight entirely.
The SaaSpocalypse heralds a structural transition from application code to AI agents managing business logic, with further changes anticipated as companies adapt to this evolving landscape in the coming years.
Keywords: #phi4, AI agents, Anthropic, CRUD databases, Cowork, Microsoft, SaaS, SaaSpocalypse, Satya Nadella, Service as Software (SaS), agentic AI, business logic, domain expertise, enterprise software, knowledge workers, legal tech, market selloff, outcome-based business models, plugins, pricing models, structural shift
thenewstack.io 9 days ago
https://blog.hermesloom.org/p/the-next-bubble-that-will 9 days ago
|
1813.
HN
A Horrible Conclusion
The article offers a critical examination of recent advancements in generative AI from an ethical perspective, specifically focusing on their application in security testing. While acknowledging the potential benefits of these technologies in automating bug detection and increasing vulnerability discovery rates, the author raises significant concerns about transparency and the quality of findings reported by companies such as Anthropic. The skepticism stems from a perceived lack of clarity regarding how effective AI tools are in identifying high-severity vulnerabilities. Despite recognizing AI's automation capabilities, the piece argues that these do not justify the associated costs and potential ethical issues.
The author contends that traditional security testing methods involving human researchers might be more efficient and safer compared to relying on AI. The article criticizes AI companies for misallocating resources towards AI development instead of supporting skilled professionals in the field. Consequently, it advises caution against incorporating these tools into security practices due to ethical concerns and inefficiencies.
The analysis concludes with a recommendation for continued research into the role of AI within this domain but emphasizes focusing on areas that do not present significant ethical dilemmas. Additionally, there is a call for the academic community to investigate other avenues for automated vulnerability discovery that avoid the ethical pitfalls associated with current generative AI technologies.
Keywords: #phi4, AI, Anthropic, LLM capabilities, academic research, attackers, automation, bug discovery, defenders, disclosure windows, disclosure windows Keywords: AI, due diligence, ethical violations, financial incentives, memory safety, misuse of funds, resource allocation, risk analysis, security testing, technical debt, vulnerabilities, zero days
addisoncrump.info 9 days ago
|
1862.
HN
Anthropic Spoof Website and How Senior Developers Look for New Work
Anthropic developed a satirical advertisement that depicted potential scenarios of AI-embedded advertising, ultimately showcasing their decision to avoid such practices. The ad was positively received, leading to the rapid creation of a spoof website by someone utilizing cutting-edge AI tools. This incident highlights the swift iteration capabilities afforded by these technologies and illustrates how senior developers can quickly explore and develop new work ideas with advanced AI resources. The project underscores both the creative potential and ethical considerations in leveraging AI for advertising purposes.
Keywords: #phi4, AI Tools, AI-embedded Advertising, Anthropic, Dating Site, Domain, Iterate, New Work, Plot Twist, Satirical Ad, Senior Developers, Spoof Website, Technical Keywords
goldenencounters.org 10 days ago
|
1912.
HN
Anthropic's team cut ad creation time from 30 minutes to 30 seconds
Austin Lau, a growth marketer at Anthropic, significantly enhanced his efficiency in ad creation by reducing the time required from 30 minutes to just 30 seconds using Claude Code, despite initially lacking coding experience. By following guidance from a colleague, he developed two key workflows: a Figma plugin for generating variations of ad creatives and a Google Ads copy workflow that streamlined brainstorming and refining ad copy into CSV files ready for upload. Previously, the manual creation of multiple ad variations in Figma and Google Docs was time-intensive. With Claude Code, Austin automated these tasks, saving nearly 30 minutes per creative update, which allowed him to focus on more strategic activities like conducting copy experiments.
Austin's experience underscores the potential of Claude Code for non-technical users to create custom workflows by starting with small projects and utilizing existing resources. His success has inspired other teams at Anthropic to adopt Claude Code for various tasks, including writing scripts, drafting case studies, and developing web development workflows, leading to significant time savings and increased productivity.
The role of growth marketers is evolving to include tool-building responsibilities akin to those of product managers, enabling them to achieve targets more efficiently by integrating AI into their workflows. This shift allows teams to concentrate less on repetitive tasks and more on strategic initiatives, enhancing overall effectiveness and innovation within the organization.
Keywords: #phi4, AI tools, Anthropic, Claude Code, Figma plugin, Google Ads, ad creation, automation, copy generation, growth marketer, marketing, non-technical, productivity, workflows
claude.com 10 days ago
|
1923.
HN
CCC (Claude's C Compiler) on Compiler Explorer
Claude's C Compiler (CCC) on Compiler Explorer provides a feature that allows users to send their source code and compilation output to Anthropic for analysis using a large language model (LLM). This AI tool aims to explain the code and its assembly output, offering potentially valuable insights. However, it is important to note that while LLMs can be helpful, they may also produce errors with high confidence. The data shared through this service is not utilized by Anthropic for training purposes and remains private under Compiler Explorer's Privacy Policy. Users are required to give their consent before accessing this explanation feature, ensuring transparency and control over the use of their information.
Keywords: #phi4, AI, Anthropic, Claude Explain, Claude's C Compiler, Compiler Explorer, Consent, Consent Request, Continue, Explain, Explorer, Large Language Model, Technical Keywords, Third Party, assembly output, compilation output, explain code, large language model (LLM), mistakes, privacy policy, source code, technical keywords Keywords: Compiler, third party company
godbolt.org 10 days ago
https://github.com/anthropics/claudes-c-compiler/i 10 days ago
https://github.com/anthropics/claudes-c-compiler/i 7 days ago
|
1937.
HN
What to know about the software selloff
Software stocks have faced a significant downturn driven by concerns over artificial intelligence (AI) disrupting the industry. This selloff was sparked by Anthropic's release of an AI tool capable of automating legal work, which heightened fears about AI's potential impact on major software companies such as Microsoft, Salesforce, and Adobe. The broader market also experienced pressure, particularly affecting asset managers with substantial investments in software.
Despite these challenges, analysts identify opportunities within the sector. Certain software offerings are deemed essential for business operations and may not be immediately vulnerable to AI advancements. Investors might find appealing buying prospects among companies that possess strong competitive advantages and solid valuations. However, predicting when the market will reach its lowest point remains difficult due to ongoing volatility. While AI presents a threat, some analysts argue that these fears are overstated and maintain confidence in the robust fundamentals of software companies.
Keywords: #phi4, AI models, Adobe, Advanced Micro Devices, Anthropic, Broadcom, Microsoft, Morningstar US Software Index, Nvidia, Salesforce, Software selloff, buying opportunities, competitive threat, disruptive technology, double-digit declines, fundamentals, institutional selling, legal work, licensing revenue, market moves, software stocks
www.morningstar.com 10 days ago
|
1938.
HN
Show HN: Syntux – generative UI for websites, not agents
Syntux is an innovative tool designed to automate the creation of user interfaces for websites using AI models, specifically leveraging Anthropic's Claude Sonnet 4.5. It enables users to define their desired UI appearance through hints, offering a customizable approach that bypasses traditional design methods. By allowing users to specify values and model parameters, Syntux facilitates an automated process for generating website designs, streamlining the development of visually appealing interfaces without relying on conventional agents. This tool exemplifies how AI can be harnessed to enhance efficiency in web design by providing a flexible platform that adapts to user-defined specifications.
Keywords: #phi4, GeneratedUI, Show HN, Syntux, UI, agents, anthropic, claude-sonnet-4-5, generative UI, hint, model, value, websites
www.getsyntux.com 10 days ago
|
1956.
HN
Agentic Coding and the Problem of Oracles
Yanqing Cheng's guest post explores the concept of "Agentic Coding and the Problem of Oracles," focusing on the integration of large language models (LLMs) into software development, particularly highlighted by Anthropic's creation of a C compiler with minimal human input. This achievement underscores both the potential and limitations of LLMs in handling complex tasks like compiling the Linux kernel. The post argues that while LLMs can automate many coding processes, they still depend on "oracles" or sources of truth to verify correctness. Traditional automated tests fall short for nuanced software requirements, which often rely on human judgment concerning usability, reliability, security, and reputation.
Cheng suggests that humans inherently act as implicit oracles through their judgments and experiences. By simulating specific personas, LLMs can better approximate these human oracles, aligning more closely with human-defined criteria of "good" software. However, translating human judgment into machine-readable formats is essential for enhancing agent autonomy. Despite the capabilities of LLMs in coding, reviewing, and testing, humans remain crucial in defining quality standards and ensuring that outputs meet these benchmarks. The role of humans shifts from direct code writing to understanding and specifying what constitutes "good" software within their specific contexts.
Keywords: #phi4, Agentic Coding, Anthropic, Autonomy, C Compiler, Claudes, Context Driven Testing, GCC, Human Judgment, LLMs, Oracle Specification, Oracles, Persona Simulation, Software Agents
epkconsulting.substack.com 11 days ago
|
1976.
HN
The Fall of the Nerds
Software stocks have recently suffered a significant downturn due to concerns that artificial intelligence (AI) is rendering many traditional software business models outdated, particularly impacting Software-as-a-Service (SaaS) companies like Microsoft and Salesforce. This decline stems from advancements in AI tools that enable individuals with minimal technical expertise to create functional software by simply instructing AIs using plain language—a process known as "vibe coding." These developments have led experts to reassess the nature of software engineering, which is increasingly seen as routine rather than creative.
Despite AI's growing role in automating various aspects of software development, human intervention remains necessary for addressing issues such as security vulnerabilities and technical debt within AI-generated code. This shift signifies a transformation from traditional roles that emphasized craftsmanship to those focused on managing automated processes. The broader implications of this technological evolution suggest the potential end of an era dominated by highly skilled technical professionals, heralding significant economic changes with far-reaching effects on careers, education, wealth distribution, and societal structures. This trend exemplifies how rapidly human capital can become obsolete in the face of new technologies, marking a profound shift in the software industry and beyond.
Keywords: #phi4, AI, Anthropic, SaaS, automation, coding tools, displacement, economic changes, engineers, human capital, innovation, obsolescence, software stocks, technical experts, vibe coding
www.noahpinion.blog 11 days ago
|
1999.
HN
A Horrible Conclusion
The article "A Horrible Conclusion," published on February 6, 2026, critically examines the use of generative AI in security testing, highlighting ethical concerns and questioning its practicality despite its potential for automating bug discovery. The author acknowledges that while AI tools like Anthropic's Claude can identify numerous vulnerabilities, they raise significant ethical issues and financial inefficiencies compared to traditional methods. The article argues that these tools may increase vulnerability discovery rates but do not justify their use due to the premature release of findings without adequate safeguards, potentially causing more harm than good.
The author advocates for prioritizing human researchers over AI investments in cybersecurity, viewing the latter as a misuse of resources. They call on academia to explore automated methods with fewer ethical concerns. Despite acknowledging the article's rushed nature, it maintains skepticism about the efficacy and ethics of current AI applications in this field.
Keywords: #phi4, AI, Anthropic, academic research, attackers, automation, defenders, due diligence, ethical violations, resource allocation, risk analysis, security testing, trolley problem, vulnerabilities
addisoncrump.info 11 days ago
|
2035.
HN
ESR: Comes the news that Anthropic has vibecoded a C compiler
Anthropic has developed a C compiler; however, users face difficulties accessing this information because their browsers have JavaScript disabled. The message advises enabling JavaScript or switching to a compatible browser to continue using x.com and view details about supported browsers. This guidance is provided through instructions available in the Help Center, ensuring users can access necessary resources by adjusting their browser settings accordingly.
Keywords: #phi4, Anthropic, C compiler, Help Center, JavaScript, browser, disabled, enabled, news, supported browsers, technical, vibecoded, xcom
twitter.com 11 days ago
|
2062.
HN
What I wish I knew before building a vibe coding platform
Ariel, VP of AI at Appwrite, provides insights into developing Imagine, a platform designed for vibe-coding that allows users to create production-ready web applications through prompting. The article highlights key learnings essential for building such platforms rather than offering step-by-step instructions. One critical aspect is prompt caching, which is vital for cost and time efficiency due to the reliance on long-running processes in vibe-coding platforms. Effective prompt caching can achieve high cache hit rates of 90-95%, significantly reducing costs and enhancing speed.
The article also discusses the importance of real-world architecture that goes beyond simple request-response models typically taught in tutorials. Real platforms must handle network issues, browser refreshes, and concurrent user actions without corrupting state. Implementing resumable streams and durable workflows is crucial for ensuring robustness and reliability.
In terms of technology choices, Imagine utilizes TanStack Start for its generated apps due to its support for server-side rendering, type-safety, and customization capabilities. Bun is selected as the runtime because of its speed and compatibility with TypeScript, facilitating rapid builds. Additionally, given the non-deterministic nature of generative AI, deterministic practices are emphasized. These include rebuilding projects after each generation, using Language Server Protocol (LSP) for real-time diagnostics, enforcing linting rules, and proactively providing context to mitigate unexpected behaviors and enhance code quality.
The article concludes by underscoring that foundational elements such as prompt caching, durable workflows, and determinism should be prioritized from the outset. These practices are crucial to avoid costly refactoring later on, offering valuable lessons for others aiming to build similar platforms efficiently.
Keywords: #phi4, AI, Anthropic, Appwrite, Bun, Imagine, Inngest, LLMs, Prompt caching, TanStack Start, cache hit rate, determinism, deterministic guardrails, durable workflows, observability, open-source, resumable streams, sandbox provisioning, server functions, vibe-coding
imagine.dev 12 days ago
|
2141.
HN
Irony alert: Anthropic helps UK.gov to build chatbot for job seekers
The UK government is collaborating with Anthropic to create an AI assistant that will provide job seekers with personalized career advice and help secure employment, with a pilot expected later this year—a move noted as ironic given Anthropic CEO Dario Amodei’s warnings about AI’s disruptive impact on the labour market. This announcement comes amid a broader “week of focused action” on AI by the Department for Science, Innovation and Technology, which includes commissioning British AI experts for open‑source public‑service tools, a Meta‑funded fellowship programme, AI‑driven analysis of transport infrastructure, and secure offline AI solutions for sensitive data. In parallel, DSIT is launching an AI Skills Hub offering free online courses aimed at equipping 10 million workers; accessed through personal accounts and featuring university and Hartree Centre content, the 36 free beginner courses are two‑thirds supplied by tech vendors—Amazon (11), Microsoft (8), and Google (7)—though a review of Microsoft’s “Get started with Microsoft 365 Copilot” criticized it as more advertorial than instructional. Meanwhile, the Department for Education is developing AI‑powered tutoring tools for students, to be available in schools by the end of 2027 and co‑designed with teachers.
Keywords: #gpt-oss:20b, 10 million, AI, AI training, Anthropic, DSIT, Meta, UKgov, free courses, job market, job seekers, open source, pilot, transport infrastructure, universities, video analysis
www.theregister.com 12 days ago
|
2259.
HN
Anthropic can win in consumer by being more open
No summary available (error)
sergey.substack.com 13 days ago
|
2263.
HN
The Fall of the Nerds
Software markets saw a rapid decline, with the iShares SaaS ETF shedding nearly $1 trillion after a surge of investor fear that AI tools—especially from Anthropic and similar firms—could render traditional software business models obsolete; earnings disappointments, incremental AI gains, and a new legal‑review platform from Anthropic amplified worry, pushing major SaaS names such as Microsoft, Salesforce, Oracle, Intuit, and AppLovin sharply lower and dragging the wider tech sector amid valuations approaching 2022‑crash lows, yet the overall market remains decoupled from a broader downturn. This volatility underscores how modern software firms rely on specialist engineers who charge for continuous access, a model now threatened by AI‑driven “vibe coding” tools like Claude Code that enable novices to generate comparable software from plain‑English prompts, effectively shrinking the technical skill set required; commentators like Andrej Karpathy, who moved from 80 % manual to 80 % AI‑generated agent coding in a month, and Jeff Sandquist of Walmart Global Tech highlight how routine, non‑creative engineering work is most amenable to automation, shifting the engineer’s role from code creation to oversight and maintenance of AI outputs, which still carry security flaws and technical debt that demand human refinement. While AI will not render software expertise wholly obsolete, it will reposition engineers as supervisors of AI‑generated systems, preserving some specialized skills even as AI expands its reach, a transition that could ripple through careers, wealth distribution, city organization, and national economies—suggesting that the era once celebrated as the “Revenge of the Nerds” may be nearing its end, with powerful forces drawn toward newly redistributed wells of wealth.
Keywords: #gpt-oss:20b-cloud, AI, Anthropic, Microsoft, Oracle, SaaS, Silicon Valley, agents, automation, coding, fear, iShares ETF, selloff, software, stocks
www.noahpinion.blog 13 days ago
|
2322.
HN
The fall of the nerds
Software‑sector equities collapsed by roughly $1 trillion in two days as investors reacted to a confluence of weak earnings, rapid AI‑model progress and a new legal‑review tool from Anthropic, triggering the most sizeable AI‑driven sell‑off on record; the ensuing decline in software‑as‑a‑service giants such as Microsoft, Salesforce, Oracle, Intuit and AppLovin dragged the broader tech index down, with valuations plummeting to a trough that echoes the 2022 crash, yet the dip is confined largely to the sector. The traditional model of software, built on highly skilled engineers delivering bespoke solutions, is in flux as AI‑coding tools like Claude Code empower even non‑developers to construct functional applications in hours, mirroring how industrial automation displaced master weavers; this shift is increasingly tangible, with Andrej Karpathy noting a transition from 80 % manual code writing to 80 % delegated to LLM agents, while Dina Bass and Jeff Sandquist emphasize the cumulative routine, “drudgery” of engineering that renders it especially vulnerable to automation. Contenders remain that AI‑generated code will still harbor security flaws and technical debt, ensuring humans will continue to fulfill critical oversight, maintenance, and debugging duties, though perhaps now as “factory” managers of AI tools rather than code artisans. Historically, new technologies can rapidly erase particular skill sets, raising the possibility that software could see a profound transformation, whereas other engineering and scientific fields may lag or evolve differently, prompting broader speculation that the current surge of technical expertise—once a driver of wealth and urban organization—may soon reach an economic inflection point that reshapes careers, education, and power dynamics.
Keywords: #gpt-oss:20b-cloud, AI, Anthropic, Bloomberg, LLM, SaaS, Silicon Valley, automation, coding, engineers, iShares ETF, software, tech debt
www.noahpinion.blog 13 days ago
|
2340.
HN
Expensively Quadratic: The LLM Agent Cost Curve
Large‑language‑model coding agents loop by sending the entire conversation back to the LLM, processing tool calls, and awaiting further user input, which leads to incremental costs for input tokens, cache writes, and output tokens, while at each step the agent writes its previous output to a prompt‑controlled cache and reads the full conversation from that cache—resulting in near‑quadratic growth in cache‑read expenses that can dominate the bill (e.g., a typical session costed ≈ $12.93, with cache‑read charges eventually making up about 87 % of total cost). An LLM gateway at exe.dev tracks token counts (not message counts) and visualizes cumulative cost against context length, with separate plots for all costs and cache‑read costs, and mouse‑over links that compare individual conversations; box plots reveal a median input of ≈ 285 tokens and output of ≈ 100 tokens, indicating substantial spread. Cost profiles differ across sessions: some incur high output costs, others high cache‑write or read costs, and cache evictions can force costly rewrites; for sessions exceeding 100 k tokens with more than 20 LLM calls, cache‑read cost scales roughly as tokens × calls rather than tokens². A simulator modeling Anthropic pricing shows input, cache write, and output costs are significantly higher (deemed “x, 1.25x, 5x” relative to cache read, which is roughly x/10), yet even with only 20 k tokens, cache reads can become the dominant cost driver. This creates a trade‑off between reducing the number of LLM calls (to keep costs low) and retaining adequate feedback loops for tool calls and iterative navigation, akin to “dead reckoning,” where agents might short‑circuit large tool outputs or spawn sub‑agents for actions such as keyword searches; similarly, a decision must be made whether to restart a new conversation or continue an existing one to balance context‑related cost and performance. The author concludes by questioning whether cost, context size, and orchestration challenges are intrinsically linked and whether Recursive Language Models could address them, inviting community perspectives.
Keywords: #gpt-oss:20b-cloud, Anthropic, LLM, Opus, agent, cache, context, conversation, cost, provider, recursive, tokens, tool
blog.exe.dev 13 days ago
|
2364.
HN
Sam Altman Responds to Anthropic Ad Campaign
A page titled “Sam Altman Responds to Anthropic Ad Campaign” contains only a message that JavaScript is disabled, advising the user to enable JavaScript or switch to a compatible browser, and providing a link to the Help Center.
Keywords: #gpt-oss:20b-cloud, Ad Campaign, Anthropic, Help Center, JavaScript, Sam Altman, browser, continue, detected, disabled, enable, supported, xcom
twitter.com 13 days ago
https://www.youtube.com/watch?v=kQRu7DdTTVA 13 days ago
https://openai.com/policies/row-terms-of-use/ 13 days ago
https://www.wsj.com/tech/ai/the-real-story-behind- 13 days ago
https://news.ycombinator.com/item?id=46894151 13 days ago
https://xcancel.com/sama/status/201913917433992818 13 days ago
https://news.ycombinator.com/item?id=46892904 13 days ago
|
2365.
HN
Show HN: I've been running OpenClaw on a $640 Mac Mini for a week. Honest report
OpenClaw is a locally‑run, always‑on AI assistant that runs on macOS, iOS, Android, Windows (WSL2), Linux, or standalone servers, and it can speak, listen, and render live canvas content while integrating with over a dozen messaging channels—including WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, Microsoft Teams, WebChat, BlueBubbles, Matrix, and Zalo—through a simple “gateway” daemon that runs as a user service under launchd or systemd; this gateway, exposed at 127.0.0.1:18789, orchestrates RPC communication with the assistant, manages onboarding via a CLI wizard that installs and starts the daemon, and supports OAuth‑based subscriptions with an API‑key fallback; the installer requires Node ≥ 22 and is executed with `npm install -g openclaw@latest` followed by `openclaw onboard`, which configures the workspace (~/.openclaw/workspace) and channel credentials (env vars or per‑channel settings). The default model recommendation is Anthropic Pro/Max (100/200) paired with Opus 4.5 for extended context and prompt‑injection resistance, though the system accepts any model and can be updated with `openclaw doctor` and `openclaw update --channel {stable|beta|dev}`; the gateway enforces security through a DM pairing policy that codes unknown senders, which the owner can approve, and supports `dmPolicy` override to `open` for permissive channels; Tailscale integration allows a “serve” (tailnet‑only) or “funnel” (public HTTPS with password) mode that keeps the gateway bound to loopback for isolation. For development, the source can be cloned and built with pnpm (or bun) using scripts like `pnpm ui:build` and `pnpm gateway:watch` for instant reload, while optional macOS, iOS, and Android client apps extend the base gateway with menu‑bar controls, canvas surfaces, and voice‑wake features; a Docker‑sandbox option further isolates non‑main sessions, limiting execution to a curated allowance of bash, process, and file operations while forbidding browser and node interaction, thereby securing multi‑channel communication and providing a scalable, self‑contained AI assistant framework.
Keywords: #gpt-oss:20b-cloud, AI, Anthropic, Docker, Node, OAuth, OpenClaw, Security, Tailscale, gateway, launchd, npm, pnpm, skills, systemd
github.com 13 days ago
|
2390.
HN
Sam Altman: I wonder why Anthropic would go for something so clearly dishonest
Sam Altman questioned Anthropic’s motives over a venture he viewed as fraudulent, and the ensuing system alert notified the user that JavaScript was disabled in their browser, offering instructions to enable the scripting language or switch to an alternative browser to access x.com.
Keywords: #gpt-oss:20b-cloud, Anthropic, Help Center, JavaScript, Sam Altman, browser, detected, disabled, dishonest, enable, list, supported, xcom
twitter.com 13 days ago
https://news.ycombinator.com/item?id=46884883 13 days ago
https://youtu.be/De-_wQpKw0s 13 days ago
https://youtu.be/FBSam25u8O4 13 days ago
https://youtu.be/3sVD3aG_azw 13 days ago
|
2391.
HN
Anthropic Ad
The message informs that JavaScript is turned off in the current browser; enabling it or switching to a browser that supports JavaScript is required in order to access x.com.
Keywords: #gpt-oss:20b-cloud, Anthropic Ad, Help Center, JavaScript, available, browser, detected, disabled, enable, list, supported, using, xcom
twitter.com 13 days ago
https://news.ycombinator.com/item?id=46884883 13 days ago
|
2423.
HN
Anthropic's new AI tool: Next black stock market day for the software industry
Anthropic’s debut of AI‑driven tools for contract review, NDAs, compliance workflows and legal templates triggered a sharp sell‑off across the software and financial sectors, pushing shares of Adobe and Salesforce to fall about 7 %, legal‑document firms more than 10 %, and PayPal below 20 % amid weak earnings and leadership upheaval, while Bitcoin slipped to around $76,000; this turbulence reflects investors’ growing belief that AI delivers tangible productivity gains, undermining the competitiveness of traditional software and finance companies. The broader tech market suffered a $285 billion decline, with key software stocks underheavy pressure—Salesforce losing roughly half its value, Adobe down 45 %, and Microsoft falling 3 % after a 13 % slide in five days, attributed partly to higher‑than‑expected AI‑infrastructure spending and slower cloud growth—while Google’s latest AI tool provoked a sell‑off in gaming stocks, underscoring the market’s shifting dynamics around emerging AI capabilities.
Keywords: #gpt-oss:20b-cloud, AI, AI agent, Adobe, Anthropic, Bitcoin, CEO, Cowork, Google, Microsoft, PayPal, Salesforce, cloud growth, compliance documents, contracts, cryptocurrency, financial markets, gaming industry, industry, infrastructure, legal documents, legal templates, price decline, sell-off, shares, software, stock market, tool
www.heise.de 14 days ago
https://news.ycombinator.com/item?id=46876720 14 days ago
https://archive.ph/9UCNH 14 days ago
https://noyb.eu/en/pay-or-okay-tech-news-site-heisede-i 14 days ago
|
2431.
HN
Anthropic: Can I get a six pack quickly?
The YouTube page titled “Anthropic: Can I get a six pack quickly?” repeats the question “Can I get a six pack quickly?” and concludes with the standard YouTube footer, complete with links and copyright notices.
Keywords: #gpt-oss:20b-cloud, Anthropic, YouTube, advertise, creators, developers, google, nfl, pack, privacy, safety, terms, ticket
www.youtube.com 14 days ago
https://news.ycombinator.com/item?id=46884883 14 days ago
|
