AIOct 27, 2026MarkTechPost
Liquid AI Releases LFM3: 6 Open-Weight 2B-Class Multimodal Models
Liquid AI is betting that tiny models tuned for real devices can run useful agents locally, and its LFM3 family makes a credible case, with clear limits.
The file
Latest
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
AIOct 27, 2026NVIDIA Developer
Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4
NVIDIA is turning GPU sharing from a polite agreement into an enforced limit, which could quietly make a lot of expensive hardware go further.
AIOct 27, 2026The Verge
OpenAI’s math and physics claims could be historic, and handpicked access is slowing their verification
OpenAI says its systems resolved the Kakeya conjecture in all dimensions, disproved the exponential time hypothesis and produced unreleased Yang-Mills physics work. Researchers take the claims seriously and object to letting the company choose who sees them first.
AIOct 27, 2026WIRED
Anthropic’s biology ambitions meet the caution of its own IPO filing
Dario Amodei has said AI could help cure most diseases within five to 10 years, and Anthropic promoted a Claude-discovered enzyme likened to Crispr. Its S-1 is far more careful: drug discovery is slow, costly and may never earn revenue, and the enzyme is now described tentatively.
AIOct 27, 2026TechCrunch
OpenAI and Anthropic both claim their AI broke the Exponential Time Hypothesis. Here’s what that would mean
Two rival labs say their models found a faster-than-expected algorithm that overturns a 1999 pillar of complexity theory. Computer scientists are now organizing to verify it, and the way they’re doing it may matter as much as the result.
AIOct 27, 2026NVIDIA Developer
NVIDIA RTX Spark Powers Neural Rendering and AI Characters in Games
NVIDIA is betting that the next Windows gaming laptop runs on Arm, renders with neural networks, and talks back through small local models instead of the cloud.
AIOct 27, 2026TechCrunch
Genie 4 makes Google’s world model steerable in real time, and the pitch is storyboards, not reality
DeepMind’s Genie 4 adds photorealism, generated dialogue and music, mid-scene prompts and two minutes of consistency. It arrives as AMD, Amazon and Rivian race to buy world-model startups, and its known flaws are a reminder that convincing isn’t the same as correct.
AIOct 27, 2026NVIDIA Developer
Fine-Tuning NVIDIA Nemotron to Know Local Rules, Not Just Local Languages
Speaking French is not the same as knowing how a French passport office works, and NVIDIA's latest recipe targets that gap with synthetic residents and strict rules.
AIOct 27, 2026WIRED
Undisclosed GPT-4 tools in UK asylum cases draw a possible inspection
The UK borders inspector is gathering information that could lead to an inspection of Home Office AI tools that summarize asylum interviews using OpenAI’s GPT-4. Campaigners report error rates above the published 9%, and applicants are not told when the tools touch their case.
AIOct 27, 2026NVIDIA Blog
Dementia researchers aim AI at the data that never gets published
C-BRAIN, an 18-institution consortium led by Washington University’s Randall Bateman, is working with Nvidia on open co-scientist tools and federated learning over unpublished ‘dark data.’ Early case studies are fast and cheap; the big projections are estimates, not results.
AIOct 27, 2026OpenAI
Retell’s voice agents hit 100 million calls a month, by a vendor’s count
Retell says it now handles over 100 million monthly calls for 4,000-plus businesses, up from about 30 million early this year, and runs at $1.6 million in revenue per employee. It picks models on its own production benchmark. Strong signals, from an OpenAI customer story.
AIOct 27, 2026NVIDIA Developer
3 Ways to Boost MoE Inference on NVIDIA DGX Spark
NVIDIA's DGX Spark tuning guide shows big local-inference speedups are real, but the honest headline is 42.5% on real work, not 185% on a synthetic test.