True Positive Weekly #116
The most important artificial intelligence and machine learning news and articles
Hey, in this issue: training LLMs to self-detoxify their language; AI for better understanding the genome; making AI-generated code more accurate; demystifying verbatim memorization in LLMs; language models can't tell what's missing; speeding up transformers by 80% by removing self-attention; and more.
[MIT] Training LLMs to self-detoxify their language
[Google] AlphaGenome: AI for better understanding the genome
[MIT] Making AI-generated code more accurate in any language
[Stanford] Demystifying verbatim memorization in large language models
[CMU] Allie: A human-aligned chess bot
AbsenceBench: Language models can't tell what's missing
Unlocking generative AI at the edge: Speeding up transformers by 80% by removing self-attention
[Models] OLMoE: Open mixture-of-experts language models
Enjoy the newsletter? Please help us make it bigger and better by sharing it with colleagues and friends.