SanSi: A Looped Typed Decision Model for System 1.5 Thinking

Key points

• Typed decision models answer a declared question without generating text: a decision head returns a probability for each of the declared options in a single forward pass.1
• We study what lies between one pass and generated reasoning: looping, in which the same layers are recursively applied several times before one typed readout.1
• Each loop lets the model revise its hidden state before it commits to an answer, without generating a token; we call this System 1.5 thinking.1
• Used as the judge for policy optimization with reinforcement learning, without gold answers, SanSi raises the generator's F1 by 7.7 points.1

Sources (2)

• [1]SanSi: A Looped Typed Decision Model for System 1.5 Thinking
  Hugging Face Daily Papers · Oct 6, 12:00 AM

  We propose SanSi, which turns a pre-trained looped language model into a typed decision model.

  Typed decision models answer a declared question without generating text: a decision head returns a probability for each of the declared options in a single forward pass.

• [2]SanSi: A Looped Typed Decision Model for System 1.5 Thinking
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:28 AM · same content

Extractive summary: sentences quoted from the sources.