Melanie Mitchell, LLMs and World Models, Part 2 Evidence For (and Against) Emergent World Models in LLM Feb 13, 2025
[...] Conclusion The claims of emergent abstract world models in LLMs are not yet supported by strong evidence. There is some evidence of such world models arising in transformers trained on narrow domains (Othello, chess, mazes, etc.) but also evidence that their abilities arise not from human-like internal models but from large “bags of heuristics”. Moreover, the notion of “world model” itself is not rigorously defined; when considering whether an agent has a particular kind of world model, we should ask what kinds of questions such a model should be able to answer, how easy or hard it should be for the agent to get answers from the model, and to what extent we would expect that the model would allow the agent to adapt to novel situations. <https://aiguide.substack.com/p/llms-and-world-models-part-2> Qui la prima parte <https://aiguide.substack.com/p/llms-and-world-models-part-1>
