Multi-hop reasoning collapses by 1M tokens; with a learning system it doesn't

2 pointsposted 7 hours ago
by lvrzhn
(zenodo.org)

1 Comments