2023
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
More tokens and more latency per answer, so it earns its place only where a wrong answer is expensive. We do not reach for it on classification or extraction.
Instead of updating weights, the agent writes down what went wrong after a failed attempt and carries that reflection into the next attempt. The reflection is stored in natural language and used as additional context, so no gradient step is required.
@inproceedings{shinn2023reflexion,
author = {Noah Shinn and Federico Cassano and Ashwin Gopinath and Karthik Narasimhan and Shunyu Yao},
title = {Reflexion: Language Agents with Verbal Reinforcement Learning},
booktitle = {Advances in Neural Information Processing Systems 36 (NeurIPS 2023)},
year = {2023},
eprint = {2303.11366},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2303.11366}
}Ejentic did not author this paper. Credit belongs to the authors named above; the note is ours.
2023
More tokens and more latency per answer, so it earns its place only where a wrong answer is expensive. We do not reach for it on classification or extraction.
2022
The pattern behind most of the agents we build. The trace is not decoration: it is the only way to debug an agent that took a wrong turn three steps ago.
2023
The benchmark we reach for when a client asks whether an agent can be trusted with a real repository. Passing tests is a much sharper signal than "the diff looks reasonable".