2000 character limit reached
Metacognition is all you need? Using Introspection in Generative Agents to Improve Goal-directed Behavior (2401.10910v2)
Published 9 Jan 2024 in q-bio.NC and cs.AI
Abstract: Recent advances in LLMs have shown impressive capabilities in various applications, yet LLMs face challenges such as limited context windows and difficulties in generalization. In this paper, we introduce a metacognition module for generative agents, enabling them to observe their own thought processes and actions. This metacognitive approach, designed to emulate System 1 and System 2 cognitive processes, allows agents to significantly enhance their performance by modifying their strategy. We tested the metacognition module on a variety of scenarios, including a situation where generative agents must survive a zombie apocalypse, and observe that our system outperform others, while agents adapt and improve their strategies to complete tasks over time.
- Daniel Kahneman “Thinking, fast and slow” Macmillan, 2011
- “Communicative Agents for Software Development”, 2023 arXiv:2307.07924 [cs.SE]
- “Generative Agents: Interactive Simulacra of Human Behavior”, 2023 arXiv:2304.03442 [cs.HC]
- “Computational Metacognition”, 2022 arXiv:2201.12885 [cs.AI]
- “Assured Learning-enabled Autonomy: A Metacognitive Reinforcement Learning Framework”, 2021 arXiv:2103.12558 [cs.AI]
- Paul M. Krueger, Falk Lieder and Thomas L. Griffiths “Enhancing metacognitive reinforcement learning using reward structures and feedback” These authors contributed equally. In Department of Psychology, University of California Berkeley, 2022 URL: https://cocosci.princeton.edu/papers/Accelerating_Metacognitive_RL-CameraReady.pdf
- George A. Miller “The magical number seven plus or minus two: some limits on our capacity for processing information”, 1956 eprint: https://pubmed.ncbi.nlm.nih.gov/13310704/
- “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”, 2021 arXiv:2005.11401 [cs.CL]
- “Breaking Bad: Unraveling Influences and Risks of User Inputs to ChatGPT for Game Story Generation” In Interactive Storytelling Cham: Springer Nature Switzerland, 2023, pp. 285–296
- “Mistral 7B”, 2023 arXiv:2310.06825 [cs.CL]
- “Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena”, 2023 arXiv:2306.05685 [cs.CL]
- “Textbooks Are All You Need II: phi-1.5 technical report”, 2023 arXiv:2309.05463 [cs.CL]
- “Phi-2: The surprising power of small language models” https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/ In Microsoft Research Blog, 2023
- “Llama 2: Open Foundation and Fine-Tuned Chat Models”, 2023 arXiv:2307.09288 [cs.CL]
- “Mixtral of Experts”, 2024 arXiv:2401.04088 [cs.LG]
- “GPT-4 Technical Report”, 2023 arXiv:2303.08774 [cs.CL]
- “Improved Baselines with Visual Instruction Tuning” arXiv:2310.03744, 2023
- “SILO Language Models: Isolating Legal Risk In a Nonparametric Datastore”, 2023 arXiv:2308.04430 [cs.CL]
- “TinyLlama: An Open-Source Small Language Model”, 2024 arXiv:2401.02385 [cs.CL]
- “PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPU”, 2023 arXiv:2312.12456 [cs.LG]
- “Efficient Memory Management for Large Language Model Serving with PagedAttention” In Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles, 2023
- Anonymous “Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing” under review In Submitted to The Twelfth International Conference on Learning Representations, 2023 URL: https://openreview.net/forum?id=02f3mUtqnM
- Katsushi Arisaka “Grand unified theory of mind and brain-part i: Space-time approach to dynamic connectomes of c. elegans and human brains by mepmos”, 2022
- Peter Byrne, Suzanna Becker and Neil Burgess “Remembering the past and imagining the future: A neural model of spatial memory and imagery” In Psychological Review 114.2, 2007, pp. 340–375 DOI: 10.1037/0033-295X.114.2.340
- R. Sasaki, A. Anzai and D.E. Angelaki “Flexible coding of object motion in multiple reference frames by parietal cortex neurons” In Nature Neuroscience 23.8, 2020, pp. 1004–1015 DOI: 10.1038/s41593-020-0656-0
- Jason Toy “Grid cells and their potential application in AI”, 2022 arXiv:2210.12068 [q-bio.NC]
- “Vector-based navigation using grid-like representations in artificial agents” In Nature 557.7705, 2018, pp. 429–433
- Niels Leadholm, Marcus Lewis and Subutai Ahmad “Grid Cell Path Integration For Movement-Based Visual Object Recognition”, 2021 arXiv:2102.09076 [cs.AI]
- “Attention Is All You Need”, 2017 arXiv:1706.03762 [cs.CL]
- James C.R. Whittington, Joseph Warren and Tim E.J. Behrens “Relating transformers to models and neural representations of the hippocampal formation” One-sentence Summary: Transformers learn brain representations and they are algorithmically related to models of the hippocampal formation. In International Conference on Learning Representations (ICLR), 2022 URL: https://openreview.net/forum?id=B8DVo9B1YE0