FrontierMath Gets a Full Proof, AMD Opens the Training Pipeline - 2026-07-28
Four signals: AI-assisted math crossed the proof gap, AMD opened a training pipeline, and policy moved on weights...
In one minute
The details
FrontierMath now counts a second AI-assisted research result - 10/10
Takeaway: A verifier hit was not enough; the milestone came only after mathematicians turned model outputs into a full proof.
What changed: Epoch now marks the 2-adic absolute Galois group problem solved, with a full write-up by David Roe and David Turturean.
Sources: epoch.ai: Epoch AI: The 2-adic Absolute Galois Group, roed314.github.io: Full proof by David Roe and David Turturean, X/Twitter @EpochAIResearch: result announcement
AMD opens every training stage for a 2.8B-active MoE - 9/10
Takeaway: Open checkpoints help research, but AMD’s ResearchRAIL license makes Instella a training artifact, not a production default.
What changed: Instella-MoE releases 16B-total, 2.8B-active checkpoints from pretraining through RL, plus recipes, data mixtures, training code, and inference code.
Sources: rocm.blogs.amd.com: AMD ROCm Blog: Instella-MoE release, Hugging Face amd/Instella-MoE-16B-A3B-Think: Hugging Face: Instella-MoE model card, GitHub: AMD Instella-MoE, X/Twitter @PrakamyaMishra: release thread
Anthropic says test capability, not model access - 9/10
Takeaway: The split is clearer: no blanket open-weight ban, but mandatory safety tests for sufficiently capable open and closed models.
What changed: Dario Amodei’s signed statement backs chip controls and anti-distillation measures while calling non-dangerous open models a public good.
Sources: Anthropic: Our position on open-weights models, X/Twitter @AnthropicAI: policy announcement
Europe pushes high-risk AI deadlines to 2027 and 2028 - 8/10
Takeaway: European AI teams get more runway, but the compliance clock moved; it did not disappear, and AI Office oversight expanded.
What changed: The AI Omnibus took effect July 27, extending Annex III rules to December 2027 and embedded-product rules to August 2028.
Sources: digital-strategy.ec.europa.eu: European Commission: AI Omnibus enters into force, ai-act-service-desk.ec.europa.eu: EU AI Act Service Desk: implementation timeline, eur-lex.europa.eu: EUR-Lex: AI Omnibus legislative text
Also worth knowing
ChatGPT work use crosses occupational boundaries (7/10): Workers borrow AI across job boundaries; this study maps usage, not whether those borrowed tasks improve real outcomes.
Quick feedback
Send what was useful, what to cut or rank lower, what was missed, and how the link list should change.



