OpenAI claims Navier-Stokes solution, dispute erupts; GPT-6 Astra on Bedrock; new genome map

The key stories

OpenAI says it solved the Navier-Stokes problem, one of the Millennium Prize Problems. That list is unsolved math questions with prize money. The solution came from an unpublished model. There is no peer review. Do not rely on it technically until independent checks exist.¹

The debate is less about compute and more about proofs. Agents write long justifications that humans can barely check. This shifts norms on authorship, citation, and verification. If you prove with AI, document sources, verification paths, and limits.²

Reports describe serious allegations around the milestone. OpenAI says it solved a roughly 90-year-old question in fluid dynamics. Independent confirmations are missing, the discussion is ongoing.³

GPT-6 Astra is now generally available on Amazon Bedrock. OpenAI promises stronger reasoning and judgment. That is not independently verified. If you build on AWS, you can provision Astra in Bedrock and plug it into existing workloads.⁴

OpenAI introduces ChatGPT Images 2.5. The provider reports over 3 billion images generated via ChatGPT Images and GPT-Image in the API. The version is meant to raise quality and controllability. Independent measurements are missing. If you use image prompts, expect changed outputs and retest your guardrails.⁵

DeepMind presents the AlphaGenome Atlas for the human genome. The team says it maps effects of 9 billion single-letter variants and could speed up research. Clinical relevance and accessibility still have to show up.⁶⁷

ASML ties TSMC, Samsung, and Intel closer. In parallel, Huawei works to break the dependency. That strengthens ASML’s role in chipmaking in the short term.⁸

Mistral AI raises 3 billion euros, the largest tech round in Europe according to a source. The valuation is 21 billion euros. That buys Mistral time and reach, despite a gap to US rivals.⁹¹⁰

Tools and releases

• Meta unveils Muse, a personal AI agent with access to email, calendar, payments, and health. Trust and privacy will be the acid test.¹¹
• Microsoft brings agentic security scans into US government environments under codename MDASH, with a focus on automated vulnerability analysis.¹²
• Adobe bakes Generative Media straight into the Premiere timeline for video, music, effects, and soundscapes without context switching.¹³
• Managed MLflow on SageMaker AI now syncs training metrics, evaluations, inference specs, and lineage into the Model Registry with lifecycle promotion.¹⁴
• Bedrock AgentCore can be evaluated via GitHub Actions. Deploy an agent, run tests, compute scores, and block PRs on regressions.¹⁵

Research

• Pathway describes Baby Dragon Hatchling. It is a post-Transformer architecture that works in latent space, a compressed feature space. They report new cost efficiency on ARC-AGI-1.¹⁶
• Benchmarks on SageMaker AI show this: G7 instances with NVIDIA Blackwell deliver noticeable price performance gains for two 30B MoE models over G5 and G6.¹⁷

Sources

1. On the Navier–Stokes Millennium Prize Problem (simonwillison.net)
2. What OpenAI’s latest controversy tells us about the future of math (technologyreview.com)
3. Drama swirls around OpenAI’s legendary mathematical milestone (theverge.com)
4. Take on your most ambitious work with GPT-6 Astra on Amazon Bedrock (aws.amazon.com)
5. Introducing ChatGPT Images 2.5 (simonwillison.net)
6. Google’s Atlas of the human genome could pave the way for new treatments (theverge.com)
7. AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome (deepmind.google)
8. ASML locks in TSMC, Samsung, and Intel while Huawei races to break its grip (the-decoder.com)
9. Mistral AI raises 3 billion euros in Europe's largest-ever tech funding round despite lagging behind rivals (the-decoder.com)
10. Mistral raises €3B as sovereign AI becomes big business (techcrunch.com)
11. Introducing Muse: The World’s First Personal AI Agent Built for Everyone (about.fb.com)
12. Codename MDASH brings agentic AI cybersecurity to the US government (microsoft.com)
13. Adobe is trying to make its AI generators idiot-proof in Premiere (theverge.com)
14. Govern models with MLflow and Amazon SageMaker AI Model Registry sync: Part 1 (aws.amazon.com)
15. Automated agent evaluation with Amazon Bedrock AgentCore and GitHub Actions (aws.amazon.com)
16. Pathway’s brain-inspired architecture development on Amazon SageMaker HyperPod (aws.amazon.com)
17. Benchmarking small LLM inference on SageMaker AI: G7 vs G5 and G6 (aws.amazon.com)