Models, products and AI startups, near-duplicates collapsed to the most credible source
Bought a Device for Hosting Models from CraigslistA Reddit user says they bought an unopened device on Craigslist for $4,000 to host models, and checked the seller on LinkedIn to verify identity before the meetup.
Ex-TikTok execs launch AI pose app SuperposeFormer TikTok employees Melody Chu and Jing Liu launched iOS camera app Superpose, which uses AI to generate four pose suggestions for selfies or photos and offers a match-pose feature. Launched in July, the app has over 22,000 downloads and users have generated more than 190,000 poses; it offers five free daily generations and paid packs for extra tries.
OpenShell applies formal methods to agent controlOpenShell shares lessons on applying formal methods to control large, long‑running AI agents, discussing permission‑review breakdowns and demonstrating use of the open‑source Z3 library to prove agent policy changes remain within approved bounds.
Take startups from prototype to production at Disrupt 2026TechCrunch says Disrupt 2026 will cover scaling startups from prototype to production, with speakers Adrian Macneil (Foxglove), John Mackey (MBRYONICS), and Boris Sofman (Bedrock Robotics). Register before September 25 to save up to $200 on passes.
The 2026 Inference Hardware RevolutionBy 2026 the AI focus shifted from training to inference: reasoning models, chain-of-thought outputs and agentic AI have sharply increased inference load, prompting vendors like Tensordyne to build inference chips (Napier) and Nvidia to call it an "inflection point of inference" at GTC 2026. Data centers and vendors are adjusting products and investments for continuous, high-volume inference.
OpenAI working with Anthropic and Google on AI safetyOpenAI is reportedly working with competitors Anthropic and Google DeepMind on measures to address AI safety issues, reflecting industry-wide concern over economic and security risks posed by AI.
Groups appeal EPA refusal to release AI impact assessmentsSeveral groups, including Public Employees for Environmental Responsibility (PEER), appealed after the U.S. Environmental Protection Agency refused to release its impact assessments for 'high-impact' AI tools, seeking disclosure of the agency's assessment documents.
Loss: a short satire about AI progressA short satire on Hacker News imagines AI progress from manual next-token generation to agents and sub-agents taking over, using a Universal Paperclips–style setup to explore where predictive AI might lead.
Nobel laureate Ressa urges limiting AI to preserve human agencyNobel laureate Maria Ressa told Reuters that AI should be limited now or humanity will lose its agency, emphasizing the urgency of AI governance and restrictions.
Yemen’s Houthis show growing sophistication and self-relianceThe New York Times reports Yemen’s Houthi movement has become more sophisticated and self-reliant in weaponry, indicating growth in their military capabilities and reduced external dependence.
Lawmakers debate federal AI oversightU.S. lawmakers are debating federal oversight of artificial intelligence, focusing on the scope and impact of potential regulation. The talks indicate Congress is moving toward creating policies for AI technologies.
Meta highlights cost savings from in-house AI chipsMeta touted cost-saving benefits of its latest in-house AI chips, saying one processor will run in data centers next year and another model is expected by the end of 2027. The move underscores big tech's push for custom chips to reduce operating costs.
Panel: research workspace with agent-built panesPanel is a local research workspace combining chat, files, PDFs, Markdown and Jupyter panes where an agent can read/write files and create custom viewers or apps. It requires Python 3.12+ and Claude Code for full features, runs at http://localhost:4173, and stores data in ~/Panel/panel.db and ~/Panel/workspaces. This is an early tester build with rough edges.
TotalEnergies, Mistral launch $115M AI exploration programTotalEnergies and Mistral launched a three-year AI program worth over $115 million to develop advanced models for exploration, reservoir characterization, and field development by combining subsurface data and expertise with Mistral's AI capabilities.
NeurIPS 2026 venue allocation concernsNeurIPS 2026 will have multiple venues; Sydney passes sold out in minutes. Authors were asked to pick preferred locations but are not guaranteed to present there, raising concerns about a single 'main' venue concentrating resources and unfairly distributing presentations.
Mathematicians debate AI's impact on mathematical researchThe piece covers concerns in the mathematics community about AI solving advanced problems: nearly 5,000 mathematicians, including 25 Fields medalists, signed a declaration titled “A Severe Misalignment of AI in Mathematics,” fearing AI could undermine conceptual understanding and the field's intellectual soil.
OpenAI Foundation gives $40M to UNC cancer centerThe OpenAI Foundation donated $40 million to UNC Lineberger Comprehensive Cancer Center to generate data aimed at improving personalized cancer vaccines. The grant is part of the foundation's second science program, Public Data for Health.
OpenAI backs bipartisan third-party safety assessmentsOpenAI publicly supported a bipartisan House proposal to introduce third-party safety assessments for major AI systems. The endorsement signals growing industry acceptance of external audits and could shape future federal oversight.
SimpliSafe debuts AI-enabled Video Doorbell Series 2SimpliSafe launched the Video Doorbell Series 2 with 2K resolution and dual-band Wi‑Fi, priced at $199.99. It integrates on-device AI, cloud vision, and facial recognition with the Active Guard Outdoor Protection service (from $49.99/month), allowing monitoring agents to intervene when suspicious activity is detected.
Ordewell: ordered plans of coding-agent tasksOrdewell turns one goal into an editable, typed plan of tasks where each task has its own runner, model, and mode and only completes when a unique completion marker appears. Plans can be rewritten before tokens are spent, preserve completed work, and support multiple runners (built-in Claude Code, Codex, OpenCode). Requires Node.js ≥20 and the TUI needs tmux.
AI underwriting firm raises $40M Series AAn artificial intelligence underwriting company raised $40M in a Series A round; the company name and investor details were not disclosed. The raise signals continued investor interest in applying AI to insurance underwriting.
Healey vows big AI push while facing energy needsMassachusetts Governor Maura Healey pledged a major push on AI while acknowledging energy costs. She noted a recently signed climate law to speed permitting and siting for clean energy, saying the state needs new homes, labs, data centers and manufacturing to meet demand.
AI Regulation as Anthropic's Business ModelA Hacker News thread discusses the idea of AI regulation as Anthropic's business model. The post and comments explore the concept, debating feasibility, motives, and potential impacts.
AI is changing Maryland's PIA processAI is rapidly changing Maryland's Public Information Act (PIA) procedures. At the 2026 MACo Fall Symposium, Deputy Public Access Ombudsman Alisa Bralove Scherr warned that AI can make overly broad requests, lacks human judgement and can be 'confidently wrong', and that officials' AI use itself may become a producible PIA record.
Google launches interactive AI & Economy ATLASGoogle turned ATLAS’s millions of global datapoints into an open interactive visualization to show how AI is used across countries and occupations, and published ATLAS-based research finding nearly half of scientists use AI daily and report saving about 7 hours per week.
AIUC raises $40M to audit enterprise AI agentsArtificial Intelligence Underwriting Company (AIUC) has built a SOC 2–inspired third-party audit and certification layer to help enterprises control AI agents. The startup announced a $40 million Series A led by Ribbit Capital, bringing total funding to $55 million, and lists customers such as Cursor, Lovable, Harvey and ElevenLabs.
SHADOW-50M: 44M param 19.8MB offline LLMThe author trained SHADOW-50M from scratch (44M parameters on 45B tokens). The model is 19.8 MB, runs ≈1,900 tok/s on laptop CPU, uses ternary weights and a large token fingerprint vocabulary, operates fully offline, and includes built‑in calculator and fast disk indexing; released as a proof of concept.
Knowtex launches frontier healthcare AI labKnowtex launched a frontier AI lab for healthcare to develop and rigorously evaluate clinical AI tools, initially focused on oncology and now covering over 200 specialties and 300 organizations. It says its benchmark KnowBench scores 97.99% for effort reduction in clinical admin tasks.
Agility Robotics unveils Digit 5 that works alongside peopleAgility Robotics introduced Digit 5, a humanoid robot claimed to operate alongside humans without safety fences, targeting warehouse and factory collaborative work.
Anthropic confirms Claude used in weapons and surveillanceAnthropic confirmed that Claude was used to develop military software and support intelligence operations, including radar analysis, ranking air-defense targets, and aiding electronic warfare. In one Taiwan scenario it modeled engagement envelopes for Patriot- and THAAD-class systems and evaluated targets.
The A.I. Daily