A.I. note takers are making lawyers nervous

Post by pkulak » Mon, May 11, 2026, 6:01 PM UTC

r_lee wrote: post #29

user_7832 wrote: post #14

Earlier quoted context omitted.

> If they miss a word they never do unintelligible, they just start playing madlibs based on the rest of the sentence. Imo this is the single biggest flaw of LLMs. They're great at a lot of things, but knowing when they're wrong (or don't have enough information to actually work on) is a critical flaw. IMO there's nothing structural about why they shouldn't be able to spot this and correct themselves - I suspect it's…

I don't think it's a training issue, it's simply that there's no inherent "I don't know" in the transformer architecture unless it's really like something completely unknown, otherwise the nearest neighbor will be chosen and that will be whatever sounds similar or is relevant, even if it might cause a problem

Post by skinfaxi » Mon, May 11, 2026, 6:14 PM UTC

elzbardico wrote: post #108

BeetleB wrote: post #92

Earlier quoted context omitted.

Not the person who said this, but it's easy: Grow up in a country where the government listening in is common (without any transparent due process), and it becomes second nature. And then add to that how easy it is to record phone conversations with today's phones (I've done it), it's easier on the brain to assume it's being recorded as opposed to wondering if it is. But yes, I don't care about my dirty jokes being r…

Not necessarily. Snowden has proven the US intelligence agencies can basically intercept anyone's electronic communications, and that they do this is a giant scale. And yet, we still see comments of people horrified when you said you assume as default that someone (or more likley, something) is listening. The normalcy bias is too strong.

Post by m463 » Mon, May 11, 2026, 6:22 PM UTC

gwbas1c wrote: post #47

Back when I was in college, in a fraternity, we always assumed that the phones were tapped. Specifically, we never spoke about alcohol or marijuana (now legal) on the phone. Even today, I generally assume that my phone could be tapped; even when talking with my trusted work colleagues, friends, and family. I'm extra careful about dirty jokes or "grey morality" in video conferences and email. The same applies to speak…

Post by cyanydeez » Mon, May 11, 2026, 6:24 PM UTC

feoren wrote: post #76

r_lee wrote: post #29

Earlier quoted context omitted.

I don't think it's a training issue, it's simply that there's no inherent "I don't know" in the transformer architecture unless it's really like something completely unknown, otherwise the nearest neighbor will be chosen and that will be whatever sounds similar or is relevant, even if it might cause a problem

The final output of the neural network part of an LLM is a vector with weights for every token, that is then usually softmaxed and picked from. Can we not quantify the uncertainty by looking at the distribution of weights of the top 10 options? Like we expect for a note-taking app that the top choice would be something like 98% certain, and if we see that the model gives a weight of 60% to "Russia" and 30% to "France…

Post by doctorpangloss » Mon, May 11, 2026, 6:35 PM UTC

runfreeapps wrote: post #113

In business communication of every kind, voice, text, emails, paper sticky-notes, doesn't matter. Assume someone is sharing your information with someone that you don't know is sharing it. Communicate as if you have an audience at all times. Also, I am not a lawyer, but the error rates are too high for anything transcribed to stand up in court with a good lawyer, unless corroborated by human witnesses.