An OpenAI model helped researchers find a counterexample to an 80-year-old Erdős conjecture — a concrete case of AI contributing novel mathematical discovery.
OpenAI model autonomously disproves a 1946 Erdős conjecture; Logical Intelligence's Aleph Prover formalizes the result in Lean 4 — first AI to crack a major open math problem.
ESMFold2 launches fully open with a 6.8B-sequence protein atlas, outperforming AlphaFold3 on protein complexes and validating therapeutic binders.
Karpathy joins Anthropic's Pretraining team to build a team using Claude to accelerate pretraining research — closing a model-accelerates-its-own-training loop at the frontier lab level.
NanoGPT-Bench finds coding agents including Codex and Claude Code achieve just 9.3% of human AI R&D progress, tuning hyperparams but missing algorithmic research breakthroughs.
Aleph (Logic International) aces PutnamBench, VeriSoftBench, and Verina. Fully autonomous formal verification agent achieves new SOTA across all major theorem benchmarks.
New study: expert persona prompting ('you are a physicist') no longer improves accuracy on frontier models. Baseline capability has rendered the technique obsolete — update your prompt libraries.
Yann LeCun left Meta in late 2025 and founded Paris-based AMI Labs, valued at $3.5B pre-launch, with a mission focused on world models over LLMs.
OpenAI's GPT-5.4 Pro contributed to solving an Erdős problem open for 60 years — a concrete milestone for AI-assisted frontier mathematics research.
Curated AI insights — sent when there's something worth your inbox.