About Xent.Tech

We are a team of researchers and engineers with a plan to construct a bridge to super-intelligence. We believe that LLMs already contain knowledge and capabilities well beyond what they demonstrate as chatbots and agents. Combining ideas from various fields of ML, Artificial Life, and Theoretical Computer Science, we are constructing an environment designed to extract and enhance these implicit capabilities.

RL post-training is reliant on human-generated tests, human evaluation, and human data labeling. While it is possible to push these beyond human limits, the cost and difficulty grows exponentially. The boundaries of these approaches are already being reached.

We believe that we can construct an environment for open-ended evolution. An environment of games that can become increasingly more difficult and more general, entirely without human intervention. An agent training in such an environment may be able to become truly super-intelligent.

We have built the Xent benchmark as a demonstration of the value and nature of the Xent environment. Using this environment, we are expanding our understanding and validating our approaches towards the creation of the world's first general super-intelligence.