Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters (Paper)

Name: Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters (Paper)
Uploaded: 2024-10-05T00:00:00.000Z
Duration: 53 min 1 s
Description: The paper (Google DeepMind with UC Berkeley) finds that spending extra compute at test time can beat a model about 14× larger, but only on easier questions and when the model won't be queried very often. The reviewer...

Yannic Kilcher · 53:01

The paper (Google DeepMind with UC Berkeley) finds that spending extra compute at test time can beat a model about 14× larger, but only on easier questions and when the model won't be queried very often. The reviewer...

Read the full summary on tuber