Reported about 4 hours ago
Thinking Machines Lab, led by Mira Murati and supported by $2 billion in funding, is tackling the issue of non-determinism in AI model responses. In a recent blog post, the lab discussed its efforts to create AI models that generate reproducible answers by controlling GPU kernel orchestrations, which currently introduce randomness in responses. This could not only improve reliability for enterprises but also benefit reinforcement learning processes. Murati has hinted that a new product aimed at researchers and startups will be revealed soon, while the lab plans to adopt an open research approach to enhance its transparency.
Source: YAHOO