LEV: Thinking Fast and Deep with Looped Decision Models

Key points

• Decision models answer typed questions about an input with one probability per option, computed in a single forward pass without generating text.1
• Like the fast System 1 of dual-process theory, they answer at once, but their depth is fixed at training time, so they spend the same computation on easy and hard questions.1
• LEV builds on a language model pretrained as a loop, reads the options after every loop step and trains all steps jointly, so one checkpoint can answer at any depth.1
• On three looped backbones, accuracy rises with depth on the same weights, mostly on hard questions, and a 2.6B LEV goes from 63.2% at depth one to 76.2% at depth four on JevBench.1

Sources (2)

• [1]LEV: Thinking Fast and Deep with Looped Decision Models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 09:20 PM

  We present LEV, a decision model that thinks deeper by rerunning its layers in a loop instead of generating tokens.

  Decision models answer typed questions about an input with one probability per option, computed in a single forward pass without generating text.

• [2]LEV: Thinking Fast and Deep with Looped Decision Models
  Hugging Face Daily Papers · Oct 14, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.