Machines thinking workshop wants to make AI models more consistent

There was a great deal of interest in what Mira Murati’s thinking laboratory is being built with $ 2 billion in seed funding and the All-Star team of former Openai researchers who have joined the laboratory. To one blog Published Wednesday, Murati’s research lab gave the world first look at one of his works: creating AI models with reproducible answers.

The publication of the research blog, entitled “Victory of Non -In -E at LLM Conclusions”, is trying to unravel the root cause of what it accidentally introduces to the AI ​​model answers. For example, ask Chatgpt the same question sometimes and you are likely to receive a wide range of answers. This has been largely accepted in the AI ​​community as a fact-today’s AI models are considered to be non-decisive systems-but the thinking laboratory Lab considers this as a soluble problem.

The position, compiled by Thinking Machines Horace He’s researcher, argues that the main cause of the randomness of AI models is the way in which GPU cores – the small programs running through Nvidia’s computer chips – are sewn together in the conclusions (everything that happens after Enter. He suggests that with the careful control of this orchestration layer, AI models can become more deterministic.

In addition to creating more reliable answers to businesses and scientists, he notes that taking AI models to create reproductive answers could also improve learning enhancement (RL). The RL is the AI ​​models reward process for correct answers, but if the answers are all slightly different, then the data gets a little noisy. Creating more consistent AI models could make the whole RL “smoother” process, according to HE. The Thinking Machines Lab has told investors that it plans to use RL Adjusting AI models for businessThe information previously mentioned.

Murati, a former head of Openai’s technology, said in July that the first product of Labinal Machines Lab will be presented in the coming months and will be “useful for researchers and newly established companies that develop custom models”. It is still unclear what this product is or whether it will use techniques from this research to create more reproducible answers.

Machine thought laboratory has also said that he is planning Post frequently blog postsCode and other information about his research in an attempt to “benefit the public, but also to improve our own research culture”. This post, the first in the new blog series of the company called “Connectionism”, seems to be part of this effort. Openai also committed to open research when it was founded, but the company has become more closed as it has become larger. We will see if Murati’s research lab remains true in this claim.

The research blog offers a rare look at one of Silicon Valley’s most secret AI. Although it does not reveal exactly where technology is going, it shows that the Machines thinking laboratory faces some of the biggest questions at the AI ​​research borders. The actual test is whether the Machines Lab thinking workshop can solve these problems and make products around his research to justify the $ 12 billion valuation.