Is an LLM a "condensate" of all knowledge? Explicit vs implicit knowledge debate
SturnioloSimone · x · 2026-11-05
A researcher argues LLMs can generate answers across vast domains because training data already contains the desired information — making a model a "condensate of all possible knowledge on almost any problem." David Manheim counters that it's unclear how much task knowledge is captured explicitly in training versus implicitly emerging from the combination of the task prompt and training data.
Related event: Researchers debate whether LLMs should generate interpretable GOFAI systems(3 posts)→