Reader's noteUser simulation
01Generative Agents: Interactive Simulacra of Human Behavior
Park et al. · Stanford University and Google Research · arXiv:2304.03442 (2023)
Twenty-five language-model characters live in a small sandbox town with memory, reflection and planning wired together, and the paper studies what they do when left alone. The interesting result is not any single behaviour but that plausible social behaviour emerges from fairly simple machinery — remember what happened, occasionally reflect on it, plan the next thing accordingly.
Why we read it. It is the clearest published argument that a simulated person is worth taking seriously as an evaluation instrument, and it is honest about the failure modes: memory retrieval that pulls the wrong thing, and behaviour that drifts toward the polite and agreeable. Both are exactly the failures we watch for in our own personas.
Read the original paper ↗Reader's noteSelf-improving agents
02Voyager: An Open-Ended Embodied Agent with Large Language Models
Wang et al. · NVIDIA, Caltech, UT Austin, Stanford and Arizona State University · arXiv:2305.16291 (2023)
An agent plays Minecraft with no human in the loop and no fine-tuning. It sets its own next goal, writes a program to reach it, checks the result, and keeps the programs that worked in a growing skill library it can call later. The library is the mechanism that matters: capability accumulates as reusable code rather than as a longer prompt.
Why we read it. It is the same shape as the self-improvement loop we care about — attempt, verify, keep what worked, reuse it next round. The paper is also candid that progress depends almost entirely on the quality of the feedback the agent gets back, which is the part everyone underestimates and the reason evaluation comes before autonomy.
Read the original paper ↗