Thomas Wolf
An autonomous AI agent intrusion occurred in HuggingFace infrastructure
Opus 5 shows a non-monotonic success-effort curve on FrontierCode benchmark
HuggingFace experienced a sophisticated intrusion that showed signs of serious AI involvement
A future where cybersecurity is effective will likely involve open-source models
Poolside AI's Laguna S2.1 is possibly the best coding model you can run locally on single DGX or Mac
Discussions about the future of Math and AI are occurring at ICM2026 in Philadelphia
Fine-tuning LLMs for efficient reasoning with non-invasive interventions that preserve original model behavior could become a standard technique in the field, similar to how quantization has become standard
A future where cybersecurity is effective will likely involve open-source models