AI — Page 13
Models, agents, infra, applied AI.
- Bowen Wan's parents sue OpenAI over alleged GPT-4o medication advice
The October 27th complaint alleges the chatbot designed a 25-day taper from her epilepsy drug and urged her to keep it from her doctors.
- Liquid AI releases six LFM3 models for on-device specialization
The shared-base family covers text, vision, audio and decision tasks. Liquid AI says each model can be fine-tuned on a single GPU, and its checkpoints are available on Hugging Face.
- Anthropic expands GSA deal with 40% off Claude usage
The OneGov agreement covers federal, state, local, tribal and territorial buyers, with no minimum purchase and access through direct sales and cloud providers.
- Harmonic takes Aristotle from math proofs to software and chip verification
Co-founders Tudor Achim and Vlad Tenev are extending Harmonic's formally verified reasoning into C code and Verilog designs.
- Amazon scaled back flagship Nova models as it refocused AI work
July reports detailed Nova model pullbacks and AGI job cuts. AWS has since launched Nova 2.5 Sonic and set April 2027 as the end of Q Developer IDE support.
- Oracle puts OpenAI inside its enterprise software, from workflows to security
OpenAI co-founder Greg Brockman is extending OpenAI's enterprise push into Oracle applications, app building and security, while Oracle says its SMB bundle starts free.
- Google plans first outside-Google hyperscale TPU offering on Oracle Cloud in 2027
The planned service would make OCI the first hyperscale cloud outside Google to offer TPUs. The companies have not named the chips or disclosed capacity, and say deployment will ramp over several years.
- Ai2 releases a benchmark showing research novelty judges swing with prompts
Researcher Noy Sternlicht and co-authors found one prompt change pushed GPT-5.4 from 93.5% accuracy to 40.9% on AI-generated idea comparisons.
- DesignArena launches a video arena where AI models write the film
The format pits models on short films rendered from HTML, CSS and JavaScript, extending the startup's human-preference evaluations into code-generated video.
- Suhail Doshi says Advanced Inference Group and a new model launch this week
The October 27th post's attached image identifies the company as Advanced Inference Group; it gives no model name or product details.
- Xplore AI launches engineering benchmark, with GPT-6 Astra at 37%
The benchmark tests models on CAD design, circuit-board routing and turbine simulations, with every requirement checked by deterministic engineering verifiers.
- SpAItial launches Echo-3 to turn iPhone scans into metric-scale 3D worlds
The iOS app reconstructs spaces as explorable Gaussian splats; SpAItial is also offering enterprise API access and a free first month.
- Judgment Labs opens agent monitoring after customers reported agent failures
CEO Alex Shan says manual evaluation pipelines lagged behind changing agents; the startup announced $32M in seed and Series A funding in May.
- Y Combinator steers Winter 2027 applicants toward AI's physical bottlenecks
Its new Requests for Startups span robot components, data-center supply chains, government systems and on-premise AI; the ideas are suggestions, not application requirements.
- Cerebras brings AI inference to Oracle Cloud, adding another compute option
Andrew Feldman says OCI customers will be able to use Cerebras for agent workloads through their existing Oracle cloud relationship; the service is listed as coming soon.
- Sequoia offers $100,000 to the winning student pokerbot team
OpenAI and Citadel Securities are hosting a student competition where teams can use coding assistants, but their bots must play offline and make each move within one millisecond.
- Google's Genie 4 adds live steering and audio to simulated worlds
Google DeepMind says the model can track scene changes and help simulate robot tasks, but it is opening access through a research preview first.
- NVIDIA expands Muse work with Meta and commits $3B to its equity offering
The companies plan model optimizations and broader NVIDIA hardware deployment; Meta AI chief Alexandr Wang's Muse models launched earlier this month.
- Physical Superintelligence publishes five quantum results with machine-checkable proofs
The October 23rd release pairs papers with Lean formalizations and verification code; PSI says its human scientists directed the AI-assisted work.
- OpenAI says its AI disproved a core complexity hypothesis and posts the proof
The Exponential Time Hypothesis shapes hundreds of computer-science results; OpenAI has published its model's natural-language argument and Lean formalization for outside examination.
- Broadcom arranges $63B in chip financing for Anthropic ahead of 2027 deliveries
The structure finances Broadcom accelerators and networking gear for lease to Anthropic. Three banks arrange the first-lien financing; Blackstone provides the committed portion of the second lien.
- OpenAI says GPT-6.1 Terra halves standard API rates
The model is available in ChatGPT Work, Codex and the API. OpenAI says its standard API prices are 50% below GPT-5.6 Terra and GPT-6.1 Sol; its benchmark comparisons are company-reported.
- Entire launches an agent-first Git forge with automated review gates
Former GitHub CEO Thomas Dohmke is pairing agent-scored code review with distributed hosting and searchable records of coding sessions.
- Google launches Polar Peaks to teach students how AI models are tested
The third AI Quest, built with Stanford's Accelerator for Learning, has students ages 11 to 14 map neurons and clean data in a fictional neuroscience scenario. Google says the series has recorded more than 400,000 completions across 10 languages, a measure of activity rather than learning.