GPT Reviews
Nvidia's CEO claims that even free AI chips from competitors can't beat Nvidia's GPUs, highlighting the company's dominance in the AI industry.
The energy demands of AI and cryptocurrency are discussed, with concerns raised about the potential consequences of their growing electricity needs.
OpenAI's Superalignment team has developed a powerful new tool called Transformer Debugger, which allows researchers to investigate specific behaviors of small language models.
Three AI research papers are discussed, including a model-stealing attack on black-box production language models, a study questioning the effectiveness of cosine-similarity in measuring semantic similarity, and the latest version of SPLADE, a library for ranking natural language queries in information retrieval systems.ย
Contact:ย ย sergi@earkind.com
Timestamps:
00:34 Introduction
01:49ย Jensen Huang says even free AI chips from his competitors can't beat Nvidia's GPUs
03:42ย The Obscene Energy Demands of A.I.
05:03ย Transformer Debugger
06:27 Fake sponsor
08:07ย Stealing Part of a Production Language Model
09:37ย Is Cosine-Similarity of Embeddings Really About Similarity?
11:05ย SPLADE-v3: New baselines for SPLADE
12:46 Outro