Verifying Solutions with LLMs
Subbarao discusses the potential of using symbolic and learned verifiers in conjunction with large language models (LLMs) for solution verification. He highlights the effectiveness of generating code for verifiers and the differences in success rates between various approaches, such as generating Python programs versus direct solution outputs. The conversation raises intriguing questions about the complexity of solutions versus code and the implications for neural networks in problem-solving.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Subbarao Kambhampati - Do o1 models search?
Related Questions