Neurodigest for the Week (#98)

Neurodigest for the Week (#98) LLM - Ramanujan-1 — A startup by one of the authors of the transformer released an 8B model trained on AMD/TPU. It performs well on general tasks but excels in agent coding, showing phenomenal results for its 8B. - Devstral 2 — local vibe coding — Mistral has redeemed itself with Large 3. Devstral Small 2 (24B) is on par with GLM 4.6 in coding and runs locally. Along with the models, they released Vibe CLI — their competitor to Claude Code. Other - Yann LeCun's departure from Meta — One of the godfathers of AI is leaving to found a startup focused on World Models. A few of my thoughts, analysis of the reasons, and the end of an era at FAIR. Personal - Looking for an Executive Assistant for my startup GenPeach.ai