Graph of Thoughts: Solving Elaborate Problems with Large Language Models
2024 — arbitrary graph of LLM thoughts with merge and refinement (AAAI)
Graph-of-thought framework — thoughts as vertices, dependencies as edges; aggregation and feedback loops.
Besta, M., Blach, N., Kubicek, A., Gerstenberger, R., Podstawski, M., Gianinazzi, L., Gajda, J., Lehmann, T., Niewiadomski, H., Nyczyk, P., & Hoefler, T. (2024). Graph of Thoughts: Solving Elaborate Problems with Large Language Models. Proceedings of the AAAI Conference on Artificial Intelligence 38(16), 17682–17690; arXiv:2308.09687. DOI: 10.1609/aaai.v38i16.29720.
Why we cite it
Graph of Thoughts (GoT) is the main literature name for DAG/graph topology with merge and refinement loops — strictly more general than CoT or ToT. Closest single-model analogue to broadcast and multi-parent thought events on the log (Thinking by Writing).
Claims we use
- Graph topology — vertices are thoughts; edges are dependencies.
- Feedback — refinement cycles are first-class (not a separate “Loop of Thought” brand).
Author (primary)
Maciej Besta — full author list in the original paper.
Corpus stance
Context — prompting framework and cost/quality benchmarks; we adopt vocabulary and topology, not task-specific scores.