framework
active
framework:two-stage-separation-of-rationale-generation-and-answer-inferencetwo-stage separation of rationale generation and answer inference
Architectural design principle that decouples rationale generation (stage 1) from answer inference (stage 2) in Multimodal-CoT.
Neighborhood — ranked by edge-count
Frameworks (1)
framework
- Multimodal-CoTimplementsA two-stage framework that separates rationale generation and answer inference by incorporating vision and language modalities.
Findings (1)
finding
- Empirical evidence that naive one-stage CoT fails in language-only setting; two-stage + vision achieves state-of-the-art.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- About chain-of-thought and process safety.
- Formalization of anchoring as posterior allocation over pattern clusters followed by output generation.
- The perspective that LLM inference decomposes into distinct computational stages, which the paper extends to looped models
- Llama's discovered algorithm: first compute integer sum via base-10 addition, then map the sum back to cyclic concept space
- Justification for why the model organism is a realistic test case for studying steering.
- Asking model to explain its own behavior after the fact when no chain-of-thought was available
- Formalization of the sense of self as a QRF constraint.
- Inference-Time Intervention: Eliciting Truthful Answers from a Language Model (Li et al., 2023)concept0.739Safety intervention that relies on activation modification, which ESR might undermine