Think More or Read More? A Token-Budget Theory of Reasoning with Shared Context

Key points

• Reasoning systems increasingly improve answers by incorporating additional context into the model's input, from multi-agent debate, where agents read their peers' responses, to self-refinement, where a model revisits its own drafts.1
• In these systems, communication consumes context as one agent's output becomes another's input.1
• We study the regime where token reading and generation draw on one fixed token window, so every token spent reading is a token that cannot be spent reasoning.1
• For multi-agent debate, the theory gives a closed-form prediction for the optimal number of agents.1

Sources (1)

• [1]Think More or Read More? A Token-Budget Theory of Reasoning with Shared Context
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 10:18 AM

  We develop a token-budget theory that formalizes this trade-off, modeling accuracy as a function of the reasoning budget remaining after reading and aggregating votes with a Condorcet jury model that permits correlated errors.

  Reasoning systems increasingly improve answers by incorporating additional context into the model's input, from multi-agent debate, where agents read their peers' responses, to self-refinement, where a model revisits its own drafts.

Extractive summary: sentences quoted from the sources.