[mvkel]

Now imagine dumping the corpus of this case into an LLM and asking it to find discrepancies. I'll bet your bollocks to a barn dance that it'd point out the extra underscore.

This is yet another example of human confirmation bias leading to terrible outcomes for other humans. Our judgment is not so virtuous.

Now the question is whether being fallible is a feature. It seems like society is interested in keeping humans in the loop FOR the possibility of mistakes; makes it feel more authentic. Meanwhile, innocent people suffer.

[phendrenad2]

There's probably a lot of alpha in getting (legal) access to a bunch of court records and finding discrepancies, and then using that to convince municipalities into buying a subscription to your discrepancy-finding AI SaaS. Give it a cool YC name like "Convictly.ai"

[mvkel]

I don't get it. We're snobs about facts being more accurate?

In a system that is innocent before guilty, more accuracy favors the person.