Mira Murati’s Thinking Machines Solves the Mystery of LLM Nondeterminism

Описание: Ex-OpenAI CTO Mira Murati and her new company Thinking Machines have just shattered the infamous LLM “reroll curse.” For years, asking ChatGPT or any large language model the same question at temperature 0 produced different answers — a nightmare for AI safety, reproducibility, and scientific research. The culprit? Lack of batch invariance in operations like MatMul, Attention, and RMSNorm. Murati’s team built batch-invariant kernels that guarantee the same answer every time, even across thousands of runs. This breakthrough isn’t about smarter models — it’s about reliable AI you can trust. The new arms race? Speed vs Determinism.

💡Subscribe for more insights on AI, innovation, and India’s global tech rise.
------------------------------------------------------------------------------------------------------------------------------------------------------------------
Check Analytics India Magazine Website - https://analyticsindiamag.com/
Download our App - https://play.google.com/store/apps/de...
------------------------------------------------------------------------------------------------------------------------------------------------------------------
Other Social Media Platforms to Connect With us :-
Instagram -   / aim.media.house  
Twitter -   / analyticsindiam  
Linkedin -   / analytics-india-magazine  
Facebook -   / analyticsindiamagazine  
------------------------------------------------------------------------------------------------------------------------------------------------------------------
#ThinkingMachines #LLM #MiraMurati #AI #FrontPageNews #AimTv #AimNetwork