xie-2021-icl-bayesian-citation

Summary

This records a verified publication fact: four Stanford researchers put out a 2021 paper arguing that when large language models appear to "learn" from examples given in a prompt, what's really happening is the model doing a form of Bayesian inference under the hood. The system treats this as a foundational, directly observed source it can cite when reasoning about how in-context learning works, rather than something derived from other claims.