2 ms·
Are people 'interpretable'? If you ask an art expert 'how much will this painting sell for at auction', he might reply '$450k'. And when questioned, he'll p
by londons_explore 4y ago
Are people 'interpretable'?
If you ask an art expert 'how much will this painting sell for at auction', he might reply '$450k'. And when questioned, he'll probably have a long answer about the brush strokes being more detailed than this other painting by the same artist, but it being worth less due to surface damage...
If our 'black box' ML models could give a similar long answer when asked 'why', would that solve the need? Because ChatGPT is getting close to being able to do just that...
- ketralnis 4y agoIf you tell that same art expert that it actually sold for $200k, they'll happily give you a post-hoc justification for that too. ChatGPT is equally good at that, you can ask it all sorts of "why" questions about falsehoods and it will confidently muse with the best armchair expert.