Emergent Reasoning Insights
Subbarao discusses an intriguing experiment highlighting GPT-4's proficiency in decoding the Caesar cipher, particularly the ROT13 variant. He emphasizes that while the model excels at this specific offset due to abundant training data, it struggles with others, raising questions about the nature of emergent reasoning in AI. This observation invites skepticism towards the interpretation of empirical studies in machine learning.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