Distributed training
Individuals train specialist LLMs across the world. By using
LLM-Router we can combine the specialist LLMs to create a more
powerful AI than any individual model.
We can also combine this platform approach with more traditional AI
techniques such as DPO and rejection sampling.
Distributed knowledge
A distributed approach to gathering data, where many individuals can each contribute a piece of the puzzle that out performs centralized, top-down approaches. This is the reason why Wikipedia is the best encyclopedia in the world. And we believe this is an important part of CHAI's approach to AI.
Incentives & Scale
All platforms operate best when incentives are done well. We've experimented with giving developers money, but so far the largest driver of AI submissions is the opportunity to get high quality feedback, recognition and satisfaction of building a highly popular LLM. Our scale helps us to collect the critical mass of feedback and LLMs necessary to establish great feedback loops.