5 ways to improve CoT faithfulness — AI Alignment Forum
alignmentforum.org · saved by 1 readers
In the case for CoT unfaithfulness is overstated, @nostalgebraist pointed out that reading the chain-of-thought (CoT) reasoning of models is neglecte…
In the case for CoT unfaithfulness is overstated, @nostalgebraist pointed out that reading the chain-of-thought (CoT) reasoning of models is neglecte…