Users: We just want precise, Boolean searching. Or like, maybe just pull up the case when we type in its exact name. Lexis/Westlaw: Okay, here is an AI assistant! - ThreadSky

design-law.bsky.social • 32 days ago

Users: We just want precise, Boolean searching. Or like, maybe just pull up the case when we type in its exact name.

Lexis/Westlaw: Okay, here is an AI assistant!

Comments

squeakinglyjen.bsky.social•31 days ago

I work at Lexis and the AI assist is separate from the search box. That doesn’t make the search box perfect ofc (you often get better results by filtering type of result or starting in a relevant practice module). I haven’t had an issue with case law searches though so sadness if this is a problem

squeakinglyjen.bsky.social•31 days ago

In defence of the Lexis AI it is cool to have it able to search all the resources on library - including all the practical guidance area-specific resources (which can be hundreds of practice notes). I’ve found it summarises cases well as an initial intro. But can’t do everything of course.

squeakinglyjen.bsky.social•31 days ago

What a summary tool can’t do is pull out nuanced/wider contextual points that might be a lot more interesting than the actual ratio of the case. - But agree having the main search functionality as effective as possible is super important for legal research+will feed the comments in this thread back.

design-law.bsky.social•31 days ago

I'm curious: Who is the intended market for summaries? I can't imagine they're actually useful to practicing lawyers. Is this functionality mainly aimed at law students?

squeakinglyjen.bsky.social•31 days ago

I think being able to summarise long documents into an overview is just useful for busy people - there’s several new immigration law cases (my practice area) each week and you need to keep an eye on them but it’s not feasible to digest them all - I’ve used it to see if a case is likely relevant.

squeakinglyjen.bsky.social•31 days ago

I did a piece of work recently for something where I was doing a summary and then I looked at the AI summary and for some things the AI summary was definitely better - like it got specific detail in fewer words. There are some things it can’t do - like know about wider legal debate and context.

squeakinglyjen.bsky.social•31 days ago

I think a lot of people use generative AI to do first drafts of things too because it’s often quicker to work from something. With the internal Lexis one it’s legal research capability seems to get better as time goes on - ie answers are better now for same prompts as previously.

rlheppner.bsky.social•31 days ago

And, if so, isn’t that a really bad idea?

squeakinglyjen.bsky.social•30 days ago

Hi Richard the AI tool at present does a range of things - you can upload documents for it to summarise or copy and paste text for the same as one of the things it can do. The target audience is genuinely just busy lawyers wanting a summary of something long. As an example I used it when -

squeakinglyjen.bsky.social•30 days ago

- an extremely dense EU/immigration case came out the other week to see if I needed to read the whole case. From an initial skim I couldn’t even work out if it was a new point of law or not. I did the summary and could see the gist and it wasn’t relevant to what I was working on.

debgoldendc.bsky.social•31 days ago

Jennifer, I use Lexis, and this was my response to Sarah

https://bsky.app/profile/debgoldendc.bsky.social/post/3lgdi5ale2s2l

I get that AI has its uses, but at 27 years in practice (eek!), I rarely looking for overviews. If I'm researching, it's to sort out details. And the product keeps moving away from that.

squeakinglyjen.bsky.social•30 days ago

I’m very happy to feed anything specific back to colleagues - I’ll also feed the general point/mood back. With Lexis it’s sometimes an issue of filtering to ‘cases’ on the left hand column to get the result you searched for? That this is so isn’t the dream but that might sift out other random things

debgoldendc.bsky.social•30 days ago

Usually I just need to get to Boolean searching. It's just bulky to get there.

squeakinglyjen.bsky.social•30 days ago

Yes I take this point. It’s frustrating when you need to know a really specific thing (and you wouldn’t need to look it up if you didn’t) and the result is like ‘in england and wales there are family courts’. I do think it has potential for practical assistance but it’s just one tool for that.

debgoldendc.bsky.social•31 days ago

The focus on AI flash seems to come at the expense of focus on innovation I would find actually helpful

design-law.bsky.social•31 days ago

Thanks, Jennifer. I really appreciate you taking the time to share this perspective.

I dashed off my initial post in frustration when I got that marketing email. But, as you can see, I wasn't the only one who reacted that way. As you've acknowledged, there's something deeper here.

design-law.bsky.social•31 days ago

I'd also like to make clear that I know (and I suspect that most of my colleagues who've chimed in also know) that the AI assistant is separate from the basic search box.

But the search box is, indeed, a problem. Not just on Lexis but on every legal research platform I've tried.

design-law.bsky.social•31 days ago

I've been using Lexis and Westlaw for over 20 years (just did that math, yikes). What I've seen is a steady degradation of basic search functionality over time. I used to be able to just type in a Boolean search and get Boolean results. Now I have to use various tricks and workarounds to get that.

design-law.bsky.social•31 days ago

Caveat: I usually use Westlaw (happy to talk more about why if you're interested). If Lexis is different in this regard, others can correct me. I don't use Bloomberg for anything other than dockets.

design-law.bsky.social•31 days ago

Anyway, we used to talk about this a lot on Law Twitter, so I know it's not a problem that's unique to me. Or a new one.

compliancedivision.bsky.social•32 days ago

I had a federal contracting tool hallucinate an ownership relationship between two companies this week. Sure makes OCI/debarment exclusions exciting (in the way nuclear energy can be exciting).

hurdbuss.bsky.social•32 days ago

I really don't miss law.

unjuriste.bsky.social•32 days ago

It takes time to get used to those things. I ended up growing used to using Casetext's semantic search feature for some things.

The active AI assistant on Westlaw/Casetext is really only useful as a gimmick. (I don't know if it works for eDiscovery stuff but I'm a researcher not a trial lawyer).

smbrnsn.bsky.social•32 days ago

I mean, I’m a researcher too. And I just want the kind of precise results that come from searching for the words and phrases I enter, and offering me the actual cases when I type in a case name.

I’m actually pretty knowledgeable in my area—more than an LLM—and I’d prefer to look for what I want.

rtushnet.bsky.social•32 days ago

I hate that I now have to disambiguate LLM in this context.

unjuriste.bsky.social•32 days ago

Oh yes I absolutely agree about pulling up cases—for some reason I *always* end up with the petition for rehearing denied order rather than the actual opinion.

About booleans, there's no reason why one can't have both boolean and AI-powered search as options since both can be useful at times.

unjuriste.bsky.social•32 days ago

To clarify when I refer to "AI-assisted search" I'm not talking about the chatbox but an AI-assisted semantic search (a natural language search).

smbrnsn.bsky.social•32 days ago

Natural language never worked well for me in Westlaw/Lexis. They do that thing that Google has started to do of ignoring words in the query or adding their own. In my experience, natural language creates way too much work!

unjuriste.bsky.social•32 days ago

I think the only one which I've tested that works/worked is Casetext, which I think is what's now on Westlaw. That being said it only works if you know *what* you're looking for, not merely the *subject* you're searching about, and so it only really works in order to dig out quotes.

forevernever.bsky.social•32 days ago

Oh no, are they using AI for Shepardizing now, too?

dondechert.bsky.social•32 days ago

Oh I hope not. That would be bonkers.

forevernever.bsky.social•32 days ago

in fairness, "having AI do X would be bonkers" has yet to stop anybody from having AI do X.

benjamin-knight.com•31 days ago

Yeah. A custom GPT all branded up and that, well, nobody asked for.

craigshay.bsky.social•32 days ago

"Precise, Boolean searching?"

(takes cigarette out of mouth)

"I haven't heard that name in years"

ctwrestle.bsky.social•31 days ago

Everyone needs to get ready for this. AI is the biggest bubble this country has ever seen, the massive investments will demand massive implementation. "AI assistants" and "AI Overviews" will be commonplace because there must be an ROI.

charliegermano.bsky.social•32 days ago

Yes! I gave that thing a Bluebook citation and it spouted out paragraphs of nonsense.

latognetti.bsky.social•32 days ago

It would also be nice if when I typed in a code section, it would take me to that statute instead of showing me every case that cites e.g. the Texas Civil Practice & Remedies Code.

sinycalbytch.bsky.social•32 days ago

YESSSSSSSSS

rsgat.bsky.social•31 days ago

What kills me is that even when I put the phrase in quotes, it still takes liberties. Give me dumb search!

design-law.bsky.social•31 days ago

Hear, hear!

debgoldendc.bsky.social•32 days ago

Just give me back easy access to Boolean searching. That's all I want.

debgoldendc.bsky.social•32 days ago

Bringing up the case when I input a case would be nice, but I'm willing to compromise.

elliemargolis.bsky.social•32 days ago

We were just discussing this in my class yesterday. Students rolled their eyes when I mentioned some of the AI “enhancements” on the legal research platforms. They know how useless and annoying they are.

joedudekjd.bsky.social•32 days ago

The way that Westlaw floundered when I typed “Youngstown” yesterday...

getmejoemiller.bsky.social•32 days ago

👀 !

chrisedwardsca4.bsky.social•32 days ago

It's incredible. I type in the name of a particular case that I need. And Westlaw produces only cases citing that case.

elliemargolis.bsky.social•32 days ago

Getting directly to a single source was SUCH a useful and efficient feature. Just gone. Sure, AI makes things so much easier 🙄🙄

chrisedwardsca4.bsky.social•32 days ago

It's not like those inefficiencies add up in an industry that tracks its time.

ralphesq.bsky.social•32 days ago

On multiple occasions I’ve typed in a really rare last name that is one of the parties and had the results be 29 suggested cases that don’t contain that name or anything close to it… and the 30th is the case.

chrisedwardsca4.bsky.social•32 days ago

I've done the same. I handle a fair number of family law appeals. Some of the most important cases have distinctive names -- twice. And yet...

elliemargolis.bsky.social•32 days ago

The enshittification continues apace.

joedudekjd.bsky.social•32 days ago

Were I teaching research, it would bear mentioning that the better way is to Google the case to find the volume-reporter-page cite, then type that into Westlaw. 😪

elliemargolis.bsky.social•32 days ago

Oh, I say this often. Though the enshittification of Google has also made that more difficult than it used to be.

forevernever.bsky.social•32 days ago

Google Scholar case search hasn't been enshittified… yet.

lxtruong.bsky.social•32 days ago

Yah. You put it in Google and you get AI Assistant crap

peterorlowicz.bsky.social•32 days ago

Google Scholar does not seem to have been broken quite as badly as regular search.

vinminen.bsky.social•32 days ago

jasl.bsky.social•31 days ago

Indeed, it actually was a thought decision for us to do not follow the path of AI-for-everything, but we love our current product :)

robjackel.bsky.social•31 days ago

Last week I looked up a case on westlaw. I correctly input the case title and citation. It asked me "did you mean [entirely different name but with the same cite]" and when, out of morbid curiosity, I clicked "yes," it then directed me to a generic search that did not include my case.

georginayeomans.bsky.social•32 days ago

Oh you wanted a super famous Supreme Court case? Let me give you the order granting cert.

rlheppner.bsky.social•32 days ago

Boy do I not need a protégé!

design-law.bsky.social•32 days ago

Can we also all agree that that name is.... 😬

proflouisjim.bsky.social•32 days ago

Which alternatives do you think were rejected? Ingénue? Lackey?

stemthebleeding.bsky.social•32 days ago

The beta version was Péon.

rlheppner.bsky.social•32 days ago

Intelligent Generative Online Researcher

aka IGOR

proflouisjim.bsky.social•32 days ago

The only AI that I need is Clippy.

digitalvergil.bsky.social•31 days ago

Oh no, i have seen how bad Google has gotten for just casually searching things, I can't imagine how frustrating it must be in a legal database.

Finding helpful sources was enough work in uni with standard library databaes, the additional slowdown and hallucinations of ai must be a nightmare

allena-bryanne.bsky.social•31 days ago

Is it bad that it seems like googles AI answers are closer than their actual search results lately? Ugh

shannonconder.bsky.social•32 days ago

We've been testing out all of the AI stuff that the vendors have been giving us - none of them are good. The library's stance is "if you're going to use them, know the limitations and ethical concerns." I have screenshots and demo drafts if you're interested!

m-nolan.bsky.social•32 days ago

They're all raising their subscription prices to academic libraries too by insisting new tools and added value of the product. Stop y'all we don't want it!!!

vizlegal.bsky.social•31 days ago

Preach. We still have Boolean search on our platform because when we tested out AI style search, it was, eh, not great. We are also getting regular complaints from customers about AI search changes to other platforms. But we do have accurate AI summaries of court judgments.

design-law.bsky.social•32 days ago

Omg people, I know how to use "adv." That is not the point. Blocking the next person who searchsplains this to me.

lawliberg.bsky.social•31 days ago

But searchsplaining is my job!

lawliberg.bsky.social•31 days ago

I hope you don't mind me pointing this out as it's lesser known: if you go to Preferences -> Search, there is a setting to check "Always run these as Terms and Connectors searches" for searches with AND, OR, or quotation marks. Basically defaults to adv: if you're using Boolean connectors.

design-law.bsky.social•31 days ago

I do actually know (and do) this. But I don't mind you sharing because: (1) others reading this might not; and (2) I know you well enough to assume you mean well.

design-law.bsky.social•31 days ago

And my larger point remains: It's annoying that I need to constantly be keeping track of new tricks to get what should be very basic functionality (especially when I change jobs and lose my settings).

lawliberg.bsky.social•31 days ago

100%!

davidpetraitis.bsky.social•32 days ago

Searchsplains!
😂🤣👍✅👀

jginsburg.bsky.social•32 days ago

I see Westlaw continues it's quest to make everybody question why they pay for it

elmunc.bsky.social•32 days ago

adv: is your friend!

design-law.bsky.social•32 days ago

Oh yes, I know.

altiam.bsky.social•32 days ago

Boolean logic over AI magic any day! Thoughts?

nonfinality.com•32 days ago

How about updating your state cite form to the new and radically changed style guide, which was officially adopted seven months ago and published months before that?

nonfinality.com•32 days ago

Bonus points—knowing not to use any “-Ohio-“ cite for cases issued before May 1, 2002, even when the court’s then-reporter assigned those cites. But that’s just dreaming.

heatherbussing.bsky.social•32 days ago

So much AI crap to wade through. Maybe we should all send them bills for the extra time it takes to do what we used to.

rbratspies.bsky.social•32 days ago

so much energy and water wasted for AI that is markedly LESS useful than the system it "disrupts"

heatherbussing.bsky.social•32 days ago

I have absolutely no interest in becoming an algorithm trainer and hallucination and slop checker.

I'm a writer. I do not want answers; I want better questions.

rbratspies.bsky.social•32 days ago

I hear you. And right now there is no way to turn off Copilot in Powerpoint.

heatherbussing.bsky.social•32 days ago

I hate Microsoft. You can theoretically do anything you can ever want to do. But you can't ever do it because there are 17 steps and it will mess up all your existing formatting.

goobermunch.bsky.social•32 days ago

And yet I’ve had two CLEs this year tell me I have an ethical obligation to use GenAI because it will keep client costs down….

rismith.bsky.social•32 days ago

I have an ethical obligation to be sure the cases I cite actually exist, too.

nolo93.bsky.social•32 days ago

Did either of these CLEs mention any of the ethics opinions (such as ABA Formal Opinion 512, or Florida Bar Advisory Op. 24-1, or PA/Phila Bar Joint Formal Op 2024-200) warning about the ethical risks of generative AI? Just curious.

goobermunch.bsky.social•32 days ago

I’m not sure they called them out specifically, but there was the parade of lawyers who’ve been bench slapped and the admonition to be sure the AI did everything right. But that was not what I’d call the focus of the presentation.

nolo93.bsky.social•32 days ago

Hope they mentioned that it's important to check to see if your court or judge has any local rules or standing orders dealing with the use of generative AI in preparing filings as well. Two of our local federal court judges have standing orders banning its use outright.

goobermunch.bsky.social•32 days ago

Actually, they were pretty good about “don’t use this for filings, have it draft memos instead.” I think that was the lead into the benchslap portion of the presentation…

peiseresq.bsky.social•32 days ago

Ok that’s entirely bananas, and reminds me that maybe we do need to add “no AI work product” to outside counsel’s retainer agreements

goobermunch.bsky.social•32 days ago

Yeah. These were some very nice attorneys from big firms. I’m glad my colleagues had the same response I did:

“Lol, no.”

hearsay.bsky.social•32 days ago

Good idea! Hadn't thought of that.

hearsay.bsky.social•32 days ago

Inspired by the idea so I took a stab at such a clause. Not legal advice, don't rely on this, yadda yadda.

volcommie.bsky.social•32 days ago

Yoink 👀

goobermunch.bsky.social•32 days ago

Does it prevent us from using the technology assisted review features of our e-discovery platforms to review documents? Because that’s going to get expensive…

peiseresq.bsky.social•32 days ago

Love this! I feel like firms will be less positive 😂

design-law.bsky.social•32 days ago

goobermunch.bsky.social•32 days ago

People are having fun playing 1.1 against 1.5…

hieronymous.bsky.social•31 days ago

I still don't see how ethical use of AI is even possible without violating confidentiality.

goobermunch.bsky.social•31 days ago

I can ask Westlaw’s AI for a memo about whether ERISA preempts state fraudulent transfer laws without a Rule 1.6 violation.

I probably can’t give an AI my client’s medical records and a crash report and ask for a demand letter.

Though I believe it’s possible to configure an LLM to not remember…

goobermunch.bsky.social•31 days ago

…prompt information. But I’m not even beer money certain of that.

hieronymous.bsky.social•31 days ago

That's the difficulty; it's a black box, anything fed into it might get regurgitated somewhere else, and there's no way of knowing. Anything even vaguely client-related might get spit out again who knows where.

I suppose that yes pure research might be useable -- but that's where it hallucinates!

cathygellis.bsky.social•32 days ago

Wtf...

I've done lectures to law classes telling them the exact opposite, walking them through all the ethical rules implicated.

jcerb.bsky.social•31 days ago

It’s a mess - some clients want promises to use AI to save time and money, others want promises to never use AI for anything. I choose the latter …

heatherbussing.bsky.social•32 days ago

Good grief

goobermunch.bsky.social•32 days ago

I know enough about the technology to know it will be more expensive to unscrew whatever it gives me.

That’s my ethical obligation.

heatherbussing.bsky.social•32 days ago

Exactly.

nothingsmonstrd.bsky.social•32 days ago

Do people making CLEs have an ethical obligation to give good advice?

goobermunch.bsky.social•32 days ago

Surprisingly, no.

nothingsmonstrd.bsky.social•32 days ago

Do they have a fiduciary obligation to give the advice that's most profitable for their employer if their employer is being given kickbacks by LLM companies?

nothingsmonstrd.bsky.social•32 days ago

(I don't know if OpenAI et al. are giving kickbacks to people advocating for adopting their products, but I'd rather think that than accept the fact that that so many people are so stupid.)

joeinaustralia.bsky.social•30 days ago

I remember when Google used to market a literal physical search engine you could plug into your network and do all that stuff. I never saw one in real life, but IIRC it was a yellow rack mounted box of the usual dimensions.

jjinandtonic.bsky.social•32 days ago

WL “assistant”: i understand you need district court precedent, but what if i listed the same circuit case 4 different times

forevernever.bsky.social•32 days ago

From a different circuit.

meshaiman.bsky.social•31 days ago

Tell it to google.

nicolebporter.bsky.social•32 days ago

Yes. This. So annoying.

jenebbeler.bsky.social•31 days ago

Apparently there’s a trick: if you type your search term and then -ai it will omit the AI. So, “Cicero -ai” as a term, with the space.

iainnash.bsky.social•32 days ago

After using Westlex's search, I'm frequently left wondering 'did that really famous case, which is really well known and in most textbooks' actually exist or am I hallucinating?

omrimarian.bsky.social•32 days ago

Okay, AI assistant. I'm looking for an article titled "Tax in the Cathedral" by Andrew Blair Stanek.

AI Assistant: Okay. Here is a memo from the California Department on Tax an Fee Administration on the application of tobacco taxes.

jamesftierney.bsky.social•32 days ago

Me to the WL product team: Can I please have Regular Expression matching

stmdc.bsky.social•32 days ago

Put “adv:” in front of your search. Not sure what it means, but Westlaw will run the search like a real search and it will eliminate most of the false positive hits.

jamesftierney.bsky.social•32 days ago

That is not regex matching but is definitely a good first step in constructing a query

vinminen.bsky.social•32 days ago

Westlaw regex would be a game changer and could exist alongside terms and connectors functionality. Just force users to enclose it in something like "(reg: ... )".

michaeljaylissner.com•31 days ago

We have experimented with regex search at CourtListener, but the performance is just about impossible. I can’t imagine it’s any better at other publishers.

aftd.dev•31 days ago

What's the magnitude of impossibility? Waiting 10 minutes for a query is obviously a nonstarter for a consumer experience but for professionals willing to pay the compute cost, maybe okay?

michaeljaylissner.com•31 days ago

The problem is lots of people making queries, each that ties up the cpu for some amount of time. It gets very difficult very quickly to handle that niche use case.

aftd.dev•31 days ago

Yeah, maybe not worth the effort but one option could be to bulk download the N most relevant cases (using traditional search methods), then run the regex locally via a desktop app.

AWS egress adds up but less quickly than compute.

vinminen.bsky.social•31 days ago

I'd guess it's down to a matter of which language the code for the site is written in, and, in turn, which search algorithm it natively implements for regex, some of which, IIRC, are exponential rather than linear-time algos.

aftd.dev•31 days ago

I think you can get linear-ish time if you pre-compile and put limits on regex complexity (e.g. no more than N characters, no look-aheads, etc.)

michaeljaylissner.com•31 days ago

Search engines work by keeping a list of every word in the corpus and mapping those words to a list of documents that contain them. Regex queries allow users to make queries that can match many, many words, and things get ugly fast because the number of matching words and docs can be huge.

getwired.bsky.social•31 days ago

TAP TAP TAP…

“It looks like you’re trying to perform a complex search. Would you like help with that?”

alysondecker.bsky.social•32 days ago

Pulling up statutes had become ridiculously hard. Like I don’t want cases that cite to the statute in my search results. I want the statute itself.

lundeeng.bsky.social•32 days ago

MSDOS Lexis was the best version, and it has all been downhill since. I still remember all the 2 letter commands.

izzos.us•32 days ago

I would be so much more likely to poke around and try out some of this AI stuff if they didn't constantly try to shove it to the front and make it the default.

Comments

Posting Rules

Reply