Test Time Learning
Jonas discusses the innovative approach of utilizing additional compute at test time to enhance predictions, particularly in the context of large language models. He highlights the importance of selecting the right data for effective learning and shares insights on outperforming larger models using the Pyle benchmark. The conversation delves into the implications of data distribution differences and retrieval strategies for model performance.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Jonas Hübotter (ETH) - Test Time Inference
Related Questions