- Is it possible to train only on math and logic materials and expect the model's response in math questions to be superior to general models with the same training/inference compute hardware?
- Are there non-LLM approaches to the above task with the goal of achieving a non-hallucinatory agent?
- Are there non-LLM approaches to the above task with the goal of achieving a non-hallucinatory agent?