hi everyone I'm Rosemarie Miller lawyers if you're curious about using AI technology in your everyday workflow I have Pablo arandondo the CIO and co-founder of case text here to tell us more Pablo thank you so much for joining me today thank you great to be here so Apollo for starters what is case text and also what is co-counsel right so case text is a platform for legal technology we've been around for about 10 years selling primarily to litigators or although not exclusively to litigators and we've been applying artificial intelligence in various forms to make you know tools that can make it easier to do legal research to draft briefs you know things of that nature and what co-counsel is is a suite of tools built on this new generation of AI these large language models that are causing such a fuss case text was given very early access to gbt4 which is the most powerful current AI model and we've been bringing it to law for like you know the last nine months wow so how can lawyers actually use this to to make their lives a little bit easier absolutely so I mean the important thing to understand with this technology is that there's a big difference between using it correctly and using it incorrectly and what I mean by that is you can't just go to a chat bot and ask it for legal research right it's unfortunately capable of hallucinating fake case law it's not up to date Etc what you can do though and what co-counsel does is we leverage the technology but anchor it into truth right which is to say we use something called retrieval augmented generation where you can only base your answer off the real case law that we return to the AI system as opposed to trying to answer off the cuff if that makes sense and so you know from for reasons of hallucinations for reasons of security and privacy basically everything we've done working with openai has been designed to make this something that lawyers can use whether for legal research uh data mining contract law preparing for a deposition e-discovery there's really a number of different ways you can apply it so does this work for any lawyer or is it specific kinds of lawyers so you know because law the whole thing is a castle built of language and this technology works at the fundamental level of language we found that it's amazingly flexible in terms of where it can be applied so our beta clients include in-house counsel at Fortune 50 companies but also you know small boutiques non-profits and some of the largest law firms in America as well really so what's the feedback you've been receiving from it it's been quite good I mean I think you know in the beginning there's a little bit of uh I don't know a little bit of discomfort I think language is something kind of uniquely human and so when we see computers able to to mimic it so well there's something almost a little unnerving um but I think once you get past that I think you know there was some you know concern oh are there going to be any lawyers left in five years or something like that which I think is completely overblown um but once they get you using it the response is really positive um it's just an amazing tool because so much of what a lawyer does uh requires you know reading and summarization right just re-articulating one thing is another way and so having AI that can assist not replace but assist an attorney uh you know is understandably been just very well received so I want to go back to you saying like are there going to be any lawyers left in five years so over here on the journalism side we're going through the same thing are there going to be any journalists left and I've been seeing calls to like slow down AI what is that about is is that because maybe they really are going to take our jobs wow okay so there's a lot there so the slow down AI um there's a range of different concerns that people have right it's everything from short-term job displacement to long-term you know full-on Rogue AI that goes you know takes over the world and things like that so there's a lot of different uh voices and different concerns going into the Slowdown system right but I will say I think most agree that the the short-term economic impact is going to be one of the first things we face on this right and you know I'll just speak from co-counsel from case Tech's perspective there's a lot of lawyer does that this AI cannot do right it's very powerful but there's a lot that it's incapable of doing and so we don't right now with the current technology see this as something where lawyers are going to be replaced um I think that certain tasks that lawyers do might you know be shifted around I think some of the billing paradigms might be shifted around but I don't think that they will won't be a fundamental need for human attorneys but you said with the technology now and we we see technology is advancing rapidly yeah can you say that with certainty no I mean that that's the truth anything where you're predicting into the future with this technology it's very difficult um you know much brighter Minds than mine disagree on some of the stuff as well right so it's not something where I think anyone can claim certainty um that's that's the truth so if I'm a lawyer and I want to get started with case text and using co-counsel what's my first step well I think uh you know varies a little bit if you're at a big Law Firm I think we have certain you know there's Executives that are just signed just to your firm that you've worked with for that if you're at a smaller firm I think you just get on the wait list and then we're we're having to triage people to get on because there's just a lot of demand but get on the waitlist and as soon as we can we'll get you processed and then and so not to not to send anyone to your competition what should a lawyer even look for when looking for a legal AI assist stand yeah so I mean look I think that uh you know case text was very early to this we're not going to be the only ones working in the space nor should we be uh I think you know definitely make sure that these the company is not putting you at risk with things like hallucinations right so it'll be it's tempting to just wrap a UI a user interface around the raw model and say look at all these cool things it can do but it's not really doing those things often because it's creating fake case law or fake quotes and things like that so I think the number one thing to do is make sure that you know the people you're working with know what they're doing and are using it responsibly um and you know insist on a trial you know when they're on the demo you call out the query don't let them pick the query you know things like that I think you know sort of always apply okay so kind of Switching gears a bit you all are working with open AI right could you tell me about that partnership how did it happen so um you know we we were uh working with openai or at least talking with open ad going back to their earlier model gbt3 um when they showed us gbt4 we just realized right away what a profound breakthrough had happened and getting to be uh in the vicinity of openai during that period leading up to the release of gpt4 it will always be sort of one of the most exhilarating uh and intellectually rewarding periods of My Life um they are such an amazing group of people and um you know this was something where uh you know they were putting out variants of the model and we were testing it on different legal tasks and giving them feedback we got to contribute a little bit of what's called reinforcement learning where are some of our reference attorneys got to go kind of teach the model how to do one little task a little bit better um so it was we got to collaborate with them closely which I'm will be forever grateful for um and um you know and then mainly it was about you know our job was to communicate to them look if you want to bring this to the legal domain here's how things have to be right you can't have them donate their data to train another model right lawyers aren't going to do that right it needs to have certain privacy certain security obviously you can't have hallucinations and things like that so our role was basically just being a domain expert right for our little niche and helping you know translate that to them and you know I'll never forget there were moments you know we had one client as the California Innocence Project right I mean we're talking getting innocent people out of jail right and he had this great write-up about the work he had done and how he's been using co-counsel to do it and some of the engineers had opened I read that and said this is why I got into AI like this is the kind of stuff I wanted to do so it's it's just been I'm eternally grateful for the the chance to have gotten to work with openai uh and for the work that they're doing do you fear that being too niched may be problematic in years to come because maybe I'm wrong here but in my mind when I'm thinking of like how Chad GPT works the the goal is almost for it to operate like a brain and if it's operating like a brain it knows everything which means is there actually a need for me to have like this sub sector this niche of the brain because why can't I just go to the source yeah no it's a great question and where you start to see things happen like that is you know gbt4 the bigger model is better at a lot of tasks that have been fine-tuned using smaller models so I take an earlier language model and I teach it sentiment analysis right you know is this tweet happier set I can spend all this time teaching that model and then all of a sudden gbt4 comes out of the box and is better at that than the one that I taught all the specific stuff right so there's no question that big generalizable models can sometimes exceed domain you know uh capacity that said I think you know there's so much that's unique to law it's not just about the language it's also knowing the workflows right it's knowing how these different pieces fit together that I do think they'll always be a place for sort of domain-specific applications if that makes sense but yes certainly in terms of the underlying technology these massive language models are in fact I think the state of the art even over domain-specific trained models so what's next for case text well you know I think um we've just been given the privilege of of getting to help bring GPT forward to law and I think we just want to earn that privilege every day um so a lot of that is uh education you know and going out and talking to folks a lot of people have a lot of questions about this stuff and it's new enough that we don't really have answers right so um I think uh working with firms both big and small to figure out you know how do you want to evaluate this what are the use cases you want to apply it and then when they're ready kind of helping them ramp it up um so we'll be this will keep us busy for a while so I'm very excited thank you so much for joining me today Pablo thank you so much for having me yeah