Reaction
Questioning the starting assumptions of Bayesian models
One post argues that adjusting probabilities cannot reveal a flawed starting model if its hypotheses exclude the answer.
TLDR
A post argues that some AI thinkers hedge their predictions without questioning the models behind them. It cites a paper to make the case that Bayesian updating can only shift belief among the hypotheses a model starts with. In a reply, another user asks whether a hypothesis assigned zero probability would stay at zero regardless of the data.
Combined views
519
1 Source, first seen ago
