Y’all, I remember when I thought that AI legal research and drafting tools were ok if they were used by lawyers, who would have the capacity to know if the cites were good - ThreadSky

kathryntewson.bsky.social • 70 days ago

Y’all, I remember when I thought that AI legal research and drafting tools were ok if they were used by lawyers, who would have the capacity to know if the cites were good

Reposted from Meagan Hassan

Do NOT do this!!! #aifail #aiandthelaw

The cites look very real, don’t mistake it. Worse, in the brief itself, there are lengthy expositions of the *facts* of these fake cases.

h/t: @dougthelawyer.bsky.social

Comments

dankeevy.bsky.social•70 days ago

If you're a professional, and you're tempted to do this, that's a sign that maybe you should step away and take a break?

Or ideally quit.

iquanyin3.bsky.social•70 days ago

uh oh…

sotsogm.bsky.social•70 days ago

Capacity? I'd say ethical requirement and professional-point-of-pride. I was Shepardizing cases up to the last minute I was filing or going into a courtroom, just out of terror I was going to embarrass myself quoting something that had been reversed.

sotsogm.bsky.social•70 days ago

And that's after pulling any case I was going to cite just because I'd seen enough bad headnotes and annotations (and bad cites by appellate judges, harrumph) to know that sometimes what was in a mention wasn't actually what had been held.

This lazy AI stuff is never gonna stop offending me, is it?

420attychicago.bsky.social•70 days ago

My freaking non-atty husband knows what fucking Shepardizing is! I legit told him about the first AI lawyer kerfuffle and his response was, "They didn't even Shepardize????"

justicehassan.bsky.social•70 days ago

I went into the study of AI two years ago really thinking this was going to be a revolution in how we could practice law efficiently. The more I learn, the more I’m convinced it’s a black hole of absolute suck.

scifantasy.bsky.social•70 days ago

Well, it's proving a way of winnowing out people who don't deserve to be lawyers.

420attychicago.bsky.social•70 days ago

AI is apparently DEI for bad attorneys, lol.

kathryntewson.bsky.social•70 days ago

Yeah, all it does is arrange text in ways that are statistically similar to those in its training corpus. It’s not fit for this purpose.

justicehassan.bsky.social•70 days ago

Exactly! And when we factor in deliberate misinformation being fed to chatbots by corporations trying to protect their IP, it’s entirely out of the ballpark of “reliable.”

plognark.bsky.social•70 days ago

It’s also a black hole of absolute suck for developing code.

It’s just bad for everything.

punkrockscience.bsky.social•70 days ago

I recently watched a presentation where the speaker highlighted legal citations as a great use for AI, especially for non-lawyers. I shudder to think.

tsumibito13.bsky.social•70 days ago

Now, the question is, will the attorney realize their mistake and own up, or try to have chatgpt generate the cases in question like Schwartz and LoDuca?

scifantasy.bsky.social•70 days ago

A more innocent time...

kathryntewson.bsky.social•70 days ago

Here’s the brief in question, btw

https://search.txcourts.gov/SearchMedia.aspx?MediaVersionID=24dcc486-018d-425f-b99c-6b9000e73968&coa=coa05&DT=Brief&MediaID=4ba469f2-219c-44a0-82fa-7050d5c00b42

cathyw.bsky.social•70 days ago

I would have thought the legal profession would have taken note of the case where the guy was required to write letters of apology to the judges whose names the LLM signed to fake cases?

scifantasy.bsky.social•70 days ago

What, you expect lawyers to pay attention to previous lawyers' mistakes?

cathyw.bsky.social•70 days ago

learning from other people's mistakes seems like a better course of action than learning from one's own, really. but.

ugh.

scifantasy.bsky.social•70 days ago

Have you met a lawyer?

justicehassan.bsky.social•70 days ago

I didn’t realize on my first pass through this that no one even signed the document. Wow. 🤦‍♀️

kathryntewson.bsky.social•70 days ago

Y’all, when most of the cases in a brief are hyperlinked but a handful aren’t, maybe that’s a frickin clue?

cldellow.com•70 days ago

As a non lawyer, this is what I don't get.

Vanilla tech should be able to identify if a cite is valid or not, assuming sufficient digitization of prior case law.

And if the LLM is proposing what it claims to be a valid cite, that implies the case was digitized, so a failure to find is a red flag.

kathryntewson.bsky.social•70 days ago

Well, it depends on what you mean by “valid.” And “identify,” I guess.

cldellow.com•70 days ago

I was thinking the most superficial stuff: did this case exist.

Not: are the facts correct, is it still good law, is there a better case, does it actually support my argument, etc.

In some sense, we're still in the part of LLMs where some of the gibberish is easy-ish to detect. It'll get worse.

stemthebleeding.bsky.social•70 days ago

A fairly infamous example of how AI handles case law was when AI cooked up a Supreme Court ruling regarding a traffic law case.

A recent one too, well within the ages of the internet.

It's a digital mad libs.

stemthebleeding.bsky.social•70 days ago

The problem is that AI .. actually isn't intelligent.

It knows a case cite SHOULD be there, because other legal documents it's been trained on have them. And it knows you're asking for one. And it's not going to tell you there is no case. So instead - it just makes one up to make you happy.

jtothe.bsky.social•70 days ago

Oh. Oh no. OH NO. OH JESUS NO.

I lose my shit worrying if I'm formatting my sources correctly when I submit proposals or asking questions about what I've read and if we can try implementing it at a location. This dude just trusted the goddamn Hallucinatron 500.

rmd1023.bsky.social•70 days ago

... ya know, I'm going to add that to the list of things like "count the fingers on hands in a picture" to my baseline "is this slop" checks.

warcabbitmwm.bsky.social•70 days ago

Oh no. Ohhh no.

jagsaysstupid.bsky.social•70 days ago

I’m sorry, how can you not find a valid case that says a gift does not require consideration?
Jfc.

karabaic.org•70 days ago

Last year, before I retired, I consistently liked the approach of IronClad for contracts management, where they used a combination of linguistic tools and AI models to spot inconsistencies in contracts and create them. I'm curious if they're still well-regarded?

Also curious about eDiscovery...

karabaic.org•70 days ago

Have LLM's ruined eDiscovery tools? Are tools like Relativity now hallucinating search results in corpuses?

420attychicago.bsky.social•70 days ago

I never liked AI in e-discovery. Not only did it not winnow out the irrelevant material, it seemed to actively continue to dump thousands of the same irrelevant documents into each review batch regardless of what we were "teaching" it.

kathryntewson.bsky.social•70 days ago

I've had really good results with machine learning for relevance in e-discovery, but that was non-LLM stuff.

karabaic.org•70 days ago

I'm glad to hear this, because it validates my anti-AI strategy at Zapproved. :-)

I didn't need AI to do document similarity. It was a bicycle for a fish.

nefcanuck.bsky.social•70 days ago

Every time I hear our funder espouse AI as a “time saving measure” I want to remind them that it’s my law license, not theirs (If they have one) that’s on the line if the AI spits out bad results and I rely on it 🫠

crary.bsky.social•70 days ago

When this started happening, I thought it was pure laziness. And there is some of that. But the more I get my head around it, I think it is the perverse incentives of the billable-hour model, which means the most successful lawyers are always in search of shortcuts.

kathryntewson.bsky.social•70 days ago

Well, it means that lawyers who want to minimize the amount of time they spend on a matter are always in search of shortcuts. I wouldn’t say such people are definitionally the most successful.

crary.bsky.social•70 days ago

Yeah. I still struggle how to quite define that -- but I think you know what I mean. There is a certain brand of monetarily successful lawyer who no doubt sees AI as a way to greatly increase their hours billed.

plognark.bsky.social•70 days ago

I had the same exact thought regarding AI making up fake framework and package references for code.

Like, surely no programmer would just leave those in and let them pull in unknown code from repositories, right?

I feel like such a naive little baby.

plognark.bsky.social•70 days ago

And so now malware writers are populating the most common fake software package names.

https://www.theregister.com/AMP/2024/03/28/ai_bots_hallucinate_software_packages/

Comments

Posting Rules

Reply