Which new mathematical theorem was invented full by LLM?
As of the information available, there isn't a specific mention of a new mathematical theorem invented entirely by a Large Language Model (LLM) like Llemma. However, Llemma models, which are fine-tuned for mathematical reasoning, have demonstrated strong performance on benchmarks for solving mathematical problems. These models are specifically adapted to excel in mathematical tasks through continued pre-training on datasets like Proof-pile-2, which includes scientific papers and mathematical data.
While they may not have invented a new theorem on their own, Llemma models can generate proofs and potentially discover insights or approaches in mathematical reasoning that might inspire new mathematical exploration. For groundbreaking discoveries like a completely new theorem, human collaboration or further validation by mathematicians is typically essential. If you're interested in recent developments or specific cases of LLMs contributing to mathematics, further exploration of recent research papers or publications might provide more insights.