#123 - Delete your info from ChatGPT, Google’s AI plans, EU act targets OSS, PaLM 2, writers’ strike
Last Week in AI
00:00
Advancements in Model Controllability and Benchmarking
This chapter explores the concept of model controllability through innovative strategies like task vectors and weight subtraction to optimize performance in language models. It discusses the effectiveness of Active Retrieval Augmented Generation and recent benchmarks that evaluate large language models on professional certification tests, while emphasizing the need for cautious interpretation of results. Furthermore, it highlights the findings from 'Frugal GPT,' which presents methods to reduce inference costs for models like GPT-4, paving the way for more economical use of advanced AI technologies.
Transcript
Play full episode
