LLM Collapse and Monte Carlo Tree Search

We are facing diminishing returns on expensive LLM training with internet data. There is a strong opinion that recycled synthetic data created at volume will not be the ultimate answer and could lead to a phenomenon known as model collapse; or corruption with the snake eating its own tail, so to speak.

There’s a new approach that doesn’t rely on training data for new advancements in AI at all, but a gamification technique that will supply the next break through in AI, like it did for human vs. machine games such as chess.

In this instance, articficial general intelligence, or human reasoning, is seen as a giant decision-tree problem where the multiple AI agents search multiple LLMs already in existence for the optimal answer.

#montecarlo
#simulation
#forensicsweirdo