Researchers introduce ‘Thinking with Looped Flows,’ a denoising approach to reasoning
The team claims state-of-the-art results on ARC-AGI-1 and ARC-AGI-2 among looped models—not across all models.
TLDR
The researchers behind “Thinking with Looped Flows” ask whether neural networks can learn to think by denoising, or removing noise. They describe training recurrent reasoning with local denoising objectives and claim state-of-the-art results on ARC-AGI-1 and ARC-AGI-2 specifically among looped models.
Combined views
80.4K
8 Sources, first seen 20d ago