MICHELANGELO: LONG CONTEXT EVALUATIONS BEYOND HAYSTACKS VIA LATENT STRUCTURE QUERIES

Summary from the TLDR AI newsletter

Another great Google paper that shows how to evaluate long context models. It is directionally similar to the recent work by Magic.