End-to-End Verification
The discussion delves into the complexities of end-to-end predictive models and the challenges of verifying their correctness. While the idea of using the world as a verifier is intriguing, it raises questions about the reliability of signals indicating accuracy and the costs associated with fine-tuning. Ultimately, the conversation highlights the need for problem-specific verifiers and the limitations of current approaches in achieving universal verification.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Prof. Subbarao Kambhampati - LLMs don't reason, they memorize (ICML2024 2/13)
Related Questions