-
Consistency is underrated
-
TextEvolve: Automated Program Discovery with LLMs
-
RefDive: Chrome extension for better paper reading, hover for citation reference
-
What’s new in RAG?
-
Better than Elo? Experiments show that social choice theory yields more faithful rankings for LLM leaderboards and preference modeling
-
Automated Anki: take any highlighted text, convert it to flashcards, and send it to an Anki deck with one keyboard stroke
-
Introducing a learnable temperature value into the softmax self-attention scores
-
Weak recurrent blocks improve language modeling performance
-
Adaptive skip connections improve training
-
Improving language modeling loss with multi-token prediction: experiments in multi-token prediction and the new FAIR paper