Thinking Machines Lab wants to make AI models more consistent

There has been a great interest in what Mira Murati’s DENK Machines Lab builds with its $ 2 billion in seed financing and the All-Star team of former OpenAi researchers who have joined the lab. In one Blog post Murati’s Research Lab was published on Wednesday and gave the world the first view of one of his projects: creating AI models with reproducible reactions.

The research blog post, entitled ‘Not -Determinism in LLM -Inservence’, tries to unpack what arbitrariness in AI model reactions. For example, ask Chatgpt the same question a few times and you will probably get a wide range of answers. This has largely been accepted in the AI ​​community as a fact today of AI models are considered as non-determinist systems but the laboratory for thinking machines regards this as a soluble problem.

The post, written by DENK Machines Lab Researcher Horace he, argues that the main cause of the arbitrariness of AI models is the way in which GPU core core – the small programs that run into Nvidia’s computer chips – are stitched together in inference processing (everything that happens after you are entered in chatgpt in chatgt). He suggests that by carefully controlling this layer of orchestration, it is possible to make AI models more deterministic.

In addition to creating more reliable reactions for companies and scientists, he notes that obtaining AI models to generate reproducible reactions could also improve the training of strengthening the reinforcement (RL). RL is the process of rewarding AI models for correct answers, but if the answers are all different, the data becomes a bit noisy. Creating more consistent AI model reactions could make the entire RL process “smoother”, he said. Thinking Machines Lab has told investors that it intends to use RL Adjust AI models for companiesThe previously reported information.

Murati, the former Chief Technology Officer of OpenAi, said in July that the first product of Thinking Machines Lab will be unveiled in the coming months and that it will “be useful for researchers and startups that develop adapted models.” It is still unclear what that product is, or that it will use techniques from this research to generate more reproducible reactions.

Thinking Machines Lab has also said that it intends Often publish blog postsCode and other information about his research in an attempt to ‘improve 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 that effort. OpenAi has also made an obligation to open research when it was founded, but the company has been closed as it has grown. We will see if the Murati research lab will remain true to this claim.

The research blog offers a rare glimpse in one of the most secret AI startups from Silicon Valley. Although it does not reveal exactly where the technology is going, it indicates that the Thinking Machines lab tackles some of the greatest demand at the AI ​​research border. The real test is whether Think Machines Lab can solve these problems and make products around his research to justify the valuation of $ 12 billion.