Five functional tokens can generalize across 40+ diverse visual reasoning tasks

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Each functional token is associated with an internalized visual operation, yet requires no visual supervision and remains a standard token in the tokenizer vocabulary.claim0.808
  Describes the properties of the functional token.
• ATLAS combines the strengths of agentic and latent reasoning by using a single discrete word (functional token) that serves both roles.claim0.744
  Core claim of the paper: a unified token bridges the gaps.
• A pair of query and key subcomponents distributed across attention heads performs previous-token behaviorfinding0.737
  VPD recovers an attention algorithm for attending to the previous token, distributed across multiple heads.
• Cognitive functions, including sentience, can potentially be achieved by very different systems and their disparate substrates.claim0.736
  Central multiple-realizability claim of the paper, from abstract and §2.
• Token-level supervision enables models to learn functional-token invocation from reasoning contextclaim0.735
  ATLAS author's assertion that functional tokens optimized via standard cross-entropy loss learn when and how to invoke operations from surrounding text.
• Keeping functional-token vocabulary compact minimizes perturbation to base model token distributionclaim0.730
  ATLAS design philosophy: five functional tokens suffice to abstract common visual operations without excessive disruption.
• For a given task, the number of all sequences which work is tiny by comparison with the huge number of all possible sequences; less than a trillionth of all 6 × 10^23 possible sequences actually work well enough.claim0.729
  A combinatorial argument that good sequences are astronomically rare, emphasizing the difficulty of discovery.
• In A/4, over 100 features primarily respond to the token 'the' in different contextsfinding0.728
  Demonstrates prevalence of token-in-context features and feature splitting of common tokens