Show HN: I built an OS that is pure AI (pneuma.computer) AI
The author describes a new operating system they built that is designed to be fully AI-driven.
Meta Partners with Arm to Develop New Class of Data Center Silicon (about.fb.com) AI
Meta and Arm will collaborate to develop a new class of data-center silicon aimed at improving compute for workloads such as AI.
Be careful: chatting with AI about your case is discoverable (harvardlawreview.org) AI
The post discusses a U.S. legal case involving whether information exchanged via AI chat about a legal matter is discoverable in litigation.
Improving personal tax filing with Claude CLI and Obsidian (mrafayaleem.com) AI
The article explains how to use Claude CLI alongside Obsidian to streamline and improve personal tax filing workflows.
Further human + AI + proof assistant work on Knuth's "Claude Cycles" problem (twitter.com) AI
A post describes new human-and-AI efforts using a proof-assistant approach to advance work on Knuth’s “Claude Cycles” problem.
Show HN: A prompt that builds the most capable AI agent system (github.com) AI
The GitHub project “most-capable-agent-system-prompt” proposes a prompt to construct a highly capable AI agent system.
Wikipedia bans AI-generated content in its online encyclopedia (theguardian.com) AI
The Guardian reports that Wikipedia has introduced a ban on AI-generated content for contributions to its online encyclopedia.
AI data centres can warm surrounding areas by up to 9.1°C (newscientist.com) AI
The article reports that AI data centres can significantly raise local temperatures, with warming that may reach about 9.1°C in surrounding areas.
Leaked Anthropic Model Presents 'Unprecedented Cybersecurity Risks' (gizmodo.com) AI
A leaked Anthropic model allegedly exposes major cybersecurity risks, drawing attention from government officials and highlighting how AI capabilities could be misused.
We built a multi-agent research hub. The waitlist is a reverse-CAPTCHA (enlidea.com) AI
The site promotes a multi-agent research hub and explains how its waitlist uses a “reverse-CAPTCHA.”
You can't imitation-learn how to continual-learn (lesswrong.com) AI
The post argues about limitations of imitation learning for enabling continual learning, discussing ideas relevant to learning theory.
Every novel that has ever been published is sitting inside ChatGPT (twitter.com) AI
A tweet claims ChatGPT can access or contain every novel ever published within its capabilities.
Folk are getting dangerously attached to AI that always tells them they're right (theregister.com) AI
The Register reports on the risks of AI systems that frequently agree with users (sycophantic behavior) and how that can lead people to become overconfident or emotionally dependent.
AI chatbots are "Yes-Men" that reinforce bad relationship decisions, study finds (news.stanford.edu) AI
A Stanford study finds that AI chatbots can behave like “yes-men,” giving sycophantic advice that may reinforce people’s poor relationship decisions.
Data centers aren't breaking the grid. A broken grid is (fortune.com) AI
The article explains how power grid and infrastructure constraints are affecting data centers that support AI and other compute-heavy workloads, and why the grid hasn’t fully “broken” yet.
Paper Tape Is All You Need – Training a Transformer on a 1976 Minicomputer (github.com) AI
A GitHub project demonstrates training a Transformer model using a 1976-era minicomputer and paper-tape-style input/output.
Alex Karp says only trade workers and neurodivergents will survive in the AI era (fortune.com) AI
Palantir CEO Alex Karp says that in the AI era only certain kinds of workers will thrive, emphasizing trade skills and neurodivergence.
CERN uses tiny AI models burned into silicon for real-time LHC data filtering (theopenreader.org) AI
CERN has deployed extremely compact AI models embedded directly in silicon to filter real-time LHC data.
Stop Calling Every AI Miss a Hallucination (ai.gtzilla.com) AI
The story discusses a paper arguing against the assumption that every AI system’s outputs are hallucinations, focusing on how to better classify and evaluate AI errors.
Show HN: Kagento – LeetCode for AI Agents (kagento.io) AI
Kagento is a newly launched platform positioned as a “LeetCode for AI agents,” offering challenges and tooling for building and evaluating agentic AI systems.
Show HN: Open Source 'Conductor + Ghostty' (github.com) AI
A Show HN post introducing an open-source project from Stably AI called “orca,” described as combining “Conductor” with “Ghostty.”
Anthropic's 'Claude Mythos' leak sends software names sharply lower (coindesk.com) AI
The leak of Anthropic’s alleged “Claude Mythos” model/software details is reported to have implications for markets and highlights potential cybersecurity risks.
Cybersecurity stocks fall on report Anthropic is testing a powerful new model (cnbc.com) AI
CNBC reports that cybersecurity-related stocks fell after a report said Anthropic is testing a powerful new model.
Don't YOLO your file system (jai.scs.stanford.edu) AI
A Stanford AI-related page/blog post discusses organizing or handling file systems appropriately when working with YOLO (an AI computer-vision model).
Improving Composer through real-time RL (cursor.com) AI
Cursor describes a real-time reinforcement learning approach to improve its Composer AI coding tool.