Latest / AI Tools for Practicing Lawyers / Episode 033 The AI Has Left The Building

Transcript

Episode 033 The AI Has Left The Building · AI Tools for Practicing Lawyers

• 0:03
  SPEAKER_00: Welcome to AI Tools for Practicing Lawyers.
• 0:07
  SPEAKER_00: Practical, no-nonsense guidance to help attorneys put AI power to work in their practice right now.
• 0:14
  SPEAKER_04: What happens when you put an AI pioneer with more than 40 years of experience in the same conversation with two podcast hosts who have some pretty strong opinions about artificial intelligence?
• 0:29
  SPEAKER_04: Well, as it turns out, you get a pretty lively conversation.
• 0:34
  SPEAKER_04: Our guest today is Jonathan Schaefer, a computer scientist who helped teach computers to master checkers and poker, and who now believes that much of the AI industry has gotten something fundamentally wrong.
• 0:48
  SPEAKER_04: His answer is kind, an AI product that can run entirely on your own computer without sending your confidential information to the cloud.
• 0:57
  SPEAKER_04: We also got into AI hallucinations, the possibility of rogue AI agents escaping their safeguards, and Jonathan's rather alarming prediction about what it may take before governments take AI safety seriously.
• 1:12
  SPEAKER_04: And through it all, we explored whether lawyers might actually benefit from having a powerful AI that knows nothing about the outside world, but can know an extraordinary amount about their own files.
• 1:26
  SPEAKER_04: Welcome to another episode of AI Tools for Practicing Lawyers.
• 1:31
  SPEAKER_04: I'm Ron Drescher.
• 1:33
  SPEAKER_04: I'm Theor Gardner.
• 1:35
  SPEAKER_04: And today we're joined by someone who's been working in artificial intelligence since long before most people had ever even heard the term.
• 1:44
  SPEAKER_04: Jonathan Schaefer is a computer scientist, professor at the University of Alberta, and one of the pioneers of modern artificial intelligence.
• 1:53
  SPEAKER_04: It's exciting to have him here.
• 1:55
  SPEAKER_04: He's also a co-founder of the Alberta Machine Intelligence Institute, one of Canada's leading AI research organizations.
• 2:04
  SPEAKER_04: Today, Jonathan is an entrepreneur and the inventor of Kind, an AI application designed to help people unlock the knowledge contained in their own digital files while maintaining control over their privacy.
• 2:19
  SPEAKER_04: He's also one of approximately 500 scientists who recently penned an open letter warning about serious cybersecurity risks associated with artificial intelligence.
• 2:32
  SPEAKER_04: Jonathan, welcome to AI Tools for Practicing Lawyers.
• 2:36
  SPEAKER_03: Thank you very much, Ron and Heather.
• 2:38
  SPEAKER_03: I'm delighted to be here.
• 2:39
  SPEAKER_04: So, Jonathan, before we get into some of the larger issues surrounding artificial intelligence, could you spend a few minutes telling us about kind and the other products you've developed?
• 2:51
  SPEAKER_03: Thank you very much.
• 2:52
  SPEAKER_03: There are things about AI that I would like to see that I think are non-negotiable.
• 2:57
  SPEAKER_03: I believe that AI should be private, it should be safe, it should be secure, it should be accurate.
• 3:05
  SPEAKER_03: But to me, those are non-negotiable items.
• 3:09
  SPEAKER_03: Yeah, there are things that don't have to be private.
• 3:12
  SPEAKER_03: I mean, if I write a research paper, I can put it in on the web and it's it's public.
• 3:18
  SPEAKER_03: But there's lots of things that I have that need to be private.
• 3:24
  SPEAKER_03: Like my medical records, my financial statements, my intellectual property, uh the work I do with with clients.
• 3:32
  SPEAKER_03: There are things that I need to do that require accuracy.
• 3:38
  SPEAKER_03: If I'm writing a document, say for a client, I can't afford to have sloppiness.
• 3:43
  SPEAKER_03: I can't afford to have mistakes.
• 3:45
  SPEAKER_03: I need accuracy.
• 3:46
  SPEAKER_03: I also want to deal with things that are secure and safe.
• 3:50
  SPEAKER_03: I don't want to worry about agents going rogue and hacking into other sites or taking my credit card and putting it somewhere else.
• 4:00
  SPEAKER_03: What I decided rather than talking the talk, I wanted to walk the walk.
• 4:04
  SPEAKER_03: So I built a product that's personal.
• 4:07
  SPEAKER_03: It's on your desktop.
• 4:09
  SPEAKER_03: It has no need for an internet connection.
• 4:11
  SPEAKER_03: It's not connected to Claude or to OpenAI or Gemini.
• 4:17
  SPEAKER_03: It provides safe, secure, accurate, private AI for you and your data.
• 4:25
  SPEAKER_04: You know what?
• 4:27
  SPEAKER_04: I'm gonna jump ahead in my outline because there's something that I wanted to talk about 20 minutes from now, but I want to talk about it now because your quest, your response really provoked it, which is I looked at the specs for kind, and I want to talk about how lawyers may get a benefit from it.
• 4:48
  SPEAKER_04: But one thing that really interested me was you've got three different versions of kind, as I recall.
• 4:55
  SPEAKER_04: There's the free version, and then there's the the regular old desktop version, and those aren't connected to the internet.
• 5:02
  SPEAKER_04: And then you've got a version that is connected to the internet, you call the hybrid version.
• 5:08
  SPEAKER_04: The the free and the uh and the desktop version have a six gigabyte footprint, and then the internet version has a half a gigabyte footprint.
• 5:21
  SPEAKER_04: What's that all about?
• 5:23
  SPEAKER_03: Okay, good.
• 5:24
  SPEAKER_03: So that there are two versions of kind.
• 5:27
  SPEAKER_03: The first version that came out in February this year, we call it Kind Hybrid.
• 5:31
  SPEAKER_03: Most of the processing happens on your desktop with complete privacy.
• 5:37
  SPEAKER_03: But there are some computations that are very expensive to do and you need high accuracy on.
• 5:47
  SPEAKER_03: And so we would send them to the internet to be analyzed, but in a privacy-preserving way.
• 5:53
  SPEAKER_03: So it's hybrid because most happens on your desktop, but some happens in the internet.
• 5:59
  SPEAKER_03: But between February and July, we worked out how to redress all of these computational roadblocks.
• 6:08
  SPEAKER_03: And in July, we introduced kind local.
• 6:12
  SPEAKER_03: 100% local.
• 6:14
  SPEAKER_03: Everything happens on your desktop.
• 6:17
  SPEAKER_03: The multi-trillion dollar companies in the US want you to take your data, move it to the internet, to where the AI is.
• 6:29
  SPEAKER_03: What we do with kind is we now flip the script.
• 6:32
  SPEAKER_03: We take high-performance AI and we move it to where your data is, which is on your desktop.
• 6:39
  SPEAKER_03: Now, because I'm interested in consumers, yes, I'm interested in small businesses, but my first love is to build something for consumers.
• 6:49
  SPEAKER_03: Most consumers have machines that have low amounts of memory, eight gigabytes, six gigabytes.
• 6:56
  SPEAKER_03: So kind local, yes, it has a six gigabyte footprint, uh, but that's for the 16 gigabytes for machines that have 16 gigabytes of memory.
• 7:06
  SPEAKER_03: It's a smaller footprint, it's about three or four gigabytes if you have an eight gigabyte machine.
• 7:11
  SPEAKER_03: So there's a a footprint, which is the AI and the AI model that sits on your machine that is adjusted to the capabilities of what your computer uh can do.
• 7:23
  SPEAKER_04: So so this let's talk about the six gigabyte footprint.
• 7:28
  SPEAKER_04: What's in there?
• 7:30
  SPEAKER_03: All right.
• 7:31
  SPEAKER_03: So the program, the AI that uh that um we've built sits on your machine and it analyzes all of your data, and that's not a problem, but to take the results of that analysis and turn it into text, turn it into something you can read, you know, asking a question you would like a beautiful, eloquent, grammatically correct answer.
• 7:58
  SPEAKER_03: And where the most of the storage goes to is an open source AI model that we use to compose the responses to questions.
• 8:11
  SPEAKER_03: All of the facts come from your data and only your data, nothing from the internet, only your data.
• 8:19
  SPEAKER_03: And the AI model, we use it to turn that into the actual response.
• 8:24
  SPEAKER_01: Heather and I both watched this documentary about um The AI doc or how I became an apocalyptimist.
• 8:31
  SPEAKER_04: Thank you, Heather.
• 8:32
  SPEAKER_04: You will probably have to repeat that multiple times.
• 8:35
  SPEAKER_01: I have a very good memory.
• 8:36
  SPEAKER_01: I can remember it.
• 8:37
  SPEAKER_04: And a lot of the quotes.
• 8:40
  SPEAKER_04: What intrigued me about that was they were talking about the development of LLMs, of the large language modules.
• 8:47
  SPEAKER_02: Yes.
• 8:48
  SPEAKER_04: And they talked about all of the data that got stuffed, that that got stuffed into these LLMs, you know, that grew and grew and grew as the compute ability grew and grew and grew.
• 9:00
  SPEAKER_04: And that's how we found ourselves where we are today with these AI, AI abilities.
• 9:07
  SPEAKER_04: I won't call them tools for this context.
• 9:09
  SPEAKER_04: So, my what I'm hearing from you is that the this footprint is more about process than about data.
• 9:19
  SPEAKER_04: Am I right or am I missing the mark?
• 9:22
  SPEAKER_03: You need all the data to build these models.
• 9:27
  SPEAKER_03: But the models we're using are all about creating responses as opposed to retrieving factual information.
• 9:38
  SPEAKER_01: Which seems a little more risky, doesn't it?
• 9:41
  SPEAKER_03: No, no.
• 9:42
  SPEAKER_03: Actually, the uh the factual information that's out there is the riskier one because these models, all of these LLMs, you can think of them as having been built by sucking in the entire internet.
• 9:57
  SPEAKER_03: And there's sites on the internet which are really high quality, and the information there is great.
• 10:04
  SPEAKER_03: And there's sites on the internet which are the opposite.
• 10:08
  SPEAKER_03: They're crap, or they're there for deliberate reasons or manipulative reasons, or political reasons, or selfish reasons, and the information is factually incorrect.
• 10:17
  SPEAKER_03: I mean, you can find places there that, you know, say that uh sites that say the earth is is square.
• 10:24
  SPEAKER_03: There's it's you can find sites that say the moon landing is a hoax.
• 10:28
  SPEAKER_03: All that stuff gets ingested in these models and is used for training.
• 10:33
  SPEAKER_03: And so the models that are on the internet can produce answers to questions that are just ridiculous.
• 10:43
  SPEAKER_03: And we know about this because of hallucinations, right?
• 10:47
  SPEAKER_04: One of the things I love about Chat GPT is that it is connected to the whole internet and it's got that limitless ability to reach out, search the web, find the response to my prompt, and deliver what it thinks is what I'm looking for.
• 11:05
  SPEAKER_04: All right.
• 11:06
  SPEAKER_04: And my question is well, okay, if I have a limited data set in my local LLM, what what does that allow me to do?
• 11:17
  SPEAKER_04: And what does that prevent me from doing?
• 11:19
  SPEAKER_03: Okay.
• 11:20
  SPEAKER_03: You can only get answers from your data, no external data.
• 11:27
  SPEAKER_03: So there are things that you want to use Chat GPT for.
• 11:31
  SPEAKER_03: I mean, when I'm when I'm, you know, I Googled this podcast, right?
• 11:36
  SPEAKER_03: Sorry, Googled.
• 11:36
  SPEAKER_03: I went into Chat GPT or uh I think I went into Gemini and asked about this podcast because I don't didn't know anything about it.
• 11:44
  SPEAKER_03: So obviously it's gonna provide information about it.
• 11:47
  SPEAKER_03: That's great.
• 11:48
  SPEAKER_03: I don't need privacy for doing those kinds of queries.
• 11:52
  SPEAKER_03: I wanted to learn about something to do with uh some cool science thing I read.
• 11:57
  SPEAKER_03: Sure, I go to Gemini or or Claude and ask a question and I get all this cool information.
• 12:03
  SPEAKER_03: That's fine.
• 12:04
  SPEAKER_03: It's all public and I get answers, although you know sometimes I I gotta check some of these because there are hallucinations.
• 12:11
  SPEAKER_03: Where kind is valued, is important, is when you have a data set of information that you want to keep private.
• 12:24
  SPEAKER_03: And the person there are personal things.
• 12:26
  SPEAKER_03: I have a collection to do with my medical information.
• 12:29
  SPEAKER_03: I don't want to put that into Google and say, hey, here's all my medical information.
• 12:35
  SPEAKER_03: Tell me things about my medical history or my taxes or my personal investments, right?
• 12:41
  SPEAKER_03: Or my intellectual property.
• 12:44
  SPEAKER_03: I don't want to put that there, right?
• 12:48
  SPEAKER_03: So by having it on my desktop, where my files reside naturally, I'm not moving them to the cloud, which is another security issue because I'm now trusting a third party.
• 13:01
  SPEAKER_03: I'm not using it in a tool like ChatGPT, which is another issue because it ingests it.
• 13:08
  SPEAKER_03: It's not just the data it ingests, it ingests the query.
• 13:12
  SPEAKER_03: It knows who did the query, it knows where I'm from, it knows everything.
• 13:16
  SPEAKER_03: I can do it on my desktop with absolute, complete privacy.
• 13:22
  SPEAKER_03: And the security is only as good as the security of your desktop.
• 13:30
  SPEAKER_03: And if you're gonna hand out passwords and whatever else, then there's no difference.
• 13:35
  SPEAKER_03: You everything is compromised.
• 13:37
  SPEAKER_03: Let me just give you one example.
• 13:39
  SPEAKER_03: Uh a couple weeks ago, well, I came back, I was on a two-week holiday, and I was in the Arctic.
• 13:45
  SPEAKER_03: Would you like to know what the internet connectivity is like in the Arctic?
• 13:49
  SPEAKER_03: Yes.
• 13:50
  SPEAKER_03: Not particularly good, I'm sure.
• 13:54
  SPEAKER_03: But I have kind on my computer, and I was working on a project, and I had complete high-performance AI using it to mine my data, my intellectual property that I had been accumulating for years.
• 14:10
  SPEAKER_03: And I did incredibly productive work using high-performance AI, essentially my own local LLM, without having to once use the internet.
• 14:23
  SPEAKER_04: All right.
• 14:24
  SPEAKER_04: So let's say we have a law firm that has an office on the North Pole, and they are kind subscribers.
• 14:34
  SPEAKER_04: Could you give us a couple of use cases that the law firm might be able to employ using your product?
• 14:41
  SPEAKER_03: Okay, so first, right now, kind local is for the desktop, it's for the individual.
• 14:47
  SPEAKER_03: We have in a few months coming out what's called kind shared, which allows multiple desktop or laptop computers to share files between them and only them.
• 14:59
  SPEAKER_03: It never goes to the cloud.
• 15:01
  SPEAKER_03: Is that like a peer-to-peer type of setup?
• 15:04
  SPEAKER_03: That's correct.
• 15:05
  SPEAKER_03: It's but it's only uh who you who you um decide you want to share with.
• 15:13
  SPEAKER_01: It's like the GPT OSS system where they it'll basically only that.
• 15:19
  SPEAKER_03: But the point is it's never going to the cloud, okay?
• 15:22
  SPEAKER_03: It never touches the cloud.
• 15:25
  SPEAKER_03: So now when you talk about use cases, so we we have people who use it basically for themselves.
• 15:33
  SPEAKER_03: They have their uh client files on their desktop or laptop where they do their work, and it never leaves their desktop and laptop, and they never need to have an internet connection because they don't want to put the files into the cloud and they don't want to ask questions about them that possibly one of these LLMs is going to use and train on.
• 15:56
  SPEAKER_03: So that's that's one example.
• 15:58
  SPEAKER_03: We have another example where because you can customize this by the data, they have real value in that they have customers that they want to move to different jurisdictions, and they create a collection with the customer's data, and they bring in the laws, the rules, the legislation for that one jurisdiction and add it to the collection, and then ask questions, and they only get answers for this customer's data in this jurisdiction.
• 16:31
  SPEAKER_03: And there's nothing in there that is going to pollute it.
• 16:36
  SPEAKER_03: I had a funny demo.
• 16:38
  SPEAKER_03: This was about a year and a half ago.
• 16:40
  SPEAKER_03: Sorry, funny is a wrong word, but I was asked by a legal society in Canada to give an introduction to large language models for lawyers.
• 16:50
  SPEAKER_03: So this was in Alberta.
• 16:53
  SPEAKER_03: It was the Edmonton or the Calgary Bar Association, something like that.
• 16:59
  SPEAKER_03: And foolishly, I gave a live demo.
• 17:04
  SPEAKER_03: And there was something going on with ethics at that time in the province.
• 17:08
  SPEAKER_03: And so I was showing some of the capabilities and I asked a question.
• 17:12
  SPEAKER_03: I said, you know, in the province of Alberta, is the following considered uh ethical or unethical for a lawyer to do?
• 17:23
  SPEAKER_03: And of course, when you ask something live, you're sort of rolling the dice.
• 17:27
  SPEAKER_02: That's true.
• 17:28
  SPEAKER_03: And the answer came back, and I don't remember whether it was yes or no, but it gave a reasoning.
• 17:34
  SPEAKER_03: And then when you looked at the references and the citations, even though this is the province of Alberta in Canada, the reference in the citations came from Maryland.
• 17:46
  SPEAKER_03: Which meant interesting.
• 17:48
  SPEAKER_03: I mean, this this was a perfect example of AI going wrong, even though the query, the prompt said, in the province of Alberta, is this ethical?
• 18:02
  SPEAKER_03: And here in kind, you just here's uh a client, here are the appropriate laws in that jurisdiction.
• 18:11
  SPEAKER_03: Everything has been analyzed, it's on your desktop, you ask a question, it doesn't know about Maryland, it doesn't know about other clients, it's only the data that you give it, and the answers that come back into your questions are only taken from those documents that you give it.
• 18:32
  SPEAKER_03: There is no opportunity to hallucinate some other client or some other law in some other country because it's not there.
• 18:43
  SPEAKER_03: It can't touch it.
• 18:44
  SPEAKER_01: So zero chance for hallucinated citations.
• 18:47
  SPEAKER_01: That's an excellent selling point.
• 18:49
  SPEAKER_04: Well, sure.
• 18:49
  SPEAKER_04: I mean, because you can't do the selling point, I swear.
• 18:51
  SPEAKER_04: Sure, because you're law because you because you have to load, you have to have, you have to know what your law is before before you start and you can go looking for law.
• 19:01
  SPEAKER_01: And you can upload it is like the ideas, right?
• 19:04
  SPEAKER_01: You can upload the applicable law.
• 19:06
  SPEAKER_01: Now, how about updating that?
• 19:08
  SPEAKER_01: Is it something that you are responsible for updating?
• 19:10
  SPEAKER_01: Because ours change here pretty often.
• 19:13
  SPEAKER_03: Right.
• 19:13
  SPEAKER_03: So um you have to add the add the laws to the collection.
• 19:18
  SPEAKER_03: We have a version which is not in the current release, where you uh, when you upload things, if if they're on the internet, you can have these links um automatically update themselves in the sense that, for example, we have it with uh a real estate customer that where, sorry, a real estate association where the real estate laws change periodically.
• 19:42
  SPEAKER_03: And so we had and so you can add to your collection the current real estate laws and set this so that every week it it looks at the site and and uh checks to see if anything has changed.
• 19:56
  SPEAKER_03: And if something has changed, it automatically uh updates your collection with the latest version of the uh of the of that site.
• 20:08
  SPEAKER_04: All right.
• 20:08
  SPEAKER_04: Now that confuses me a little bit because one of the benefits of the of kind is the walled garden.
• 20:17
  SPEAKER_04: Correct.
• 20:17
  SPEAKER_04: And here you are, you're introducing an internet quality to it.
• 20:22
  SPEAKER_04: Yeah, yeah, I see what you're saying, Ron.
• 20:24
  SPEAKER_01: If you're I'm just wondering how you're linking it, it's got an access point right there.
• 20:28
  SPEAKER_01: So there's right.
• 20:29
  SPEAKER_03: There's there's two issues.
• 20:30
  SPEAKER_03: There's your private data, and this information is public data.
• 20:37
  SPEAKER_03: It goes into a separate collection.
• 20:39
  SPEAKER_01: And so you're within the same poll?
• 20:43
  SPEAKER_03: It they're both in kind, but they're both separate.
• 20:46
  SPEAKER_03: And you would need you need the to make the decision that you want to connect them.
• 20:51
  SPEAKER_03: You want to say, yeah, here's my client data and it's private.
• 20:55
  SPEAKER_03: And I've created this other collection, which is the applicable laws I want to use.
• 21:00
  SPEAKER_03: And if you connect them, then we'll use both of them.
• 21:04
  SPEAKER_03: But if you don't connect them, you you don't get it.
• 21:07
  SPEAKER_03: Right?
• 21:07
  SPEAKER_03: I mean, if the laws, I have to get them from somewhere.
• 21:10
  SPEAKER_01: I see.
• 21:11
  SPEAKER_01: I think what I see, what's happening.
• 21:12
  SPEAKER_01: So the attorney could keep that separate and then just search that separate and then paste it, like keep it in a completely separate portion or wall portion so that it's not connected to their client files.
• 21:24
  SPEAKER_03: Right.
• 21:25
  SPEAKER_03: Exactly.
• 21:26
  SPEAKER_03: This is this is in a separate collection, it's analyzed completely separately.
• 21:31
  SPEAKER_03: If you want to use both of those collections to answer a question about a client, you can enable that.
• 21:38
  SPEAKER_03: Otherwise, the information of your client is completely private and separate from the laws.
• 21:46
  SPEAKER_04: You know, I appreciate the marketing that this is not uh a hallucination risk type uh activity, but candidly, John.
• 22:00
  SPEAKER_04: Jonathan, it doesn't seem like the best use cases for kind would implicate the kind of things that lead to hallucinations.
• 22:10
  SPEAKER_04: You know, I like the most uh well-known one in the legal field is made-up citations and briefs filed with the court.
• 22:19
  SPEAKER_04: But I don't get the feeling that that's the best use for kind anyway.
• 22:24
  SPEAKER_04: Well, you know, that maybe some better uses for kind, at least in the law office, would be, you know, kind of document managements.
• 22:32
  SPEAKER_03: So, uh, okay, so uh first of all, you won't get hallucinated uh citations with kind, because we have your documents, we extract all the facts, we remember for every fact which file and where it came from.
• 22:48
  SPEAKER_03: And so when we compose an answer, you get the answer with citations to the document, the video, the picture with the appropriate uh indication of where it happened, whether it be on page five of this PDF file or 18 minutes and 20 seconds into this video.
• 23:09
  SPEAKER_03: So kind does not have any citation uh issues whatsoever.
• 23:15
  SPEAKER_03: But let's think about it this way: we're using your computer.
• 23:20
  SPEAKER_03: I don't know about you, but my computer's pretty powerful.
• 23:24
  SPEAKER_03: Uh has a bunch of cores, it has a GPU or an NPU, an AI processing chip.
• 23:31
  SPEAKER_03: And even though I'm a computer scientist and I do a lot of computing, the reality is my computer's idle most of the time.
• 23:38
  SPEAKER_03: It's it's probably not running during the evening, it's not running on the weekends and holidays.
• 23:42
  SPEAKER_03: I'm going for lunch, it's not running then, right?
• 23:46
  SPEAKER_03: It's idle.
• 23:46
  SPEAKER_03: There's incredible resources you have on your desktop.
• 23:50
  SPEAKER_03: One of the features we have that resonates with lawyers is what we call consistency checking.
• 23:57
  SPEAKER_03: So if you have a document, hey, you can run a spell checker on it, great.
• 24:02
  SPEAKER_03: Fixes your spelling errors.
• 24:04
  SPEAKER_03: You can run a grammar checker on it, great.
• 24:06
  SPEAKER_03: It fixes all your grammar errors.
• 24:08
  SPEAKER_03: But we have a logic checker.
• 24:10
  SPEAKER_03: So if you have a collection, I'll give you an example.
• 24:14
  SPEAKER_03: I wrote a book.
• 24:16
  SPEAKER_03: It's it's got uh uh uh 150, 200 files in it.
• 24:19
  SPEAKER_03: Some of them are Word documents, some of them are pictures, some of their videos, et cetera.
• 24:24
  SPEAKER_03: When you do the consistency checker, it looks at the facts that's been extracted from all the files in the collection and sees if there's any inconsistencies.
• 24:36
  SPEAKER_03: And of course, I ran it before I published my book earlier this year, and it found some logic mistakes.
• 24:41
  SPEAKER_03: I said X here, I said Y there.
• 24:44
  SPEAKER_03: Some of them were factually incorrect.
• 24:47
  SPEAKER_03: I had mistakes in my book, logic mistakes, fact mistakes, and I fixed them.
• 24:52
  SPEAKER_03: And so one of the things that lawyers have found useful is you've got a collection, say, of all of these documents to do with a particular lawyer or particular class of documents in a jurisdiction, whatever it is.
• 25:06
  SPEAKER_03: And the consistency checker goes through and makes sure that everything is internally consistent, that there is nothing that is factually incorrect.
• 25:12
  SPEAKER_03: And the reason why I'm pointing that out is because that's not a cheap computation.
• 25:17
  SPEAKER_03: Now you can send it all this stuff to the cloud and get it done, but what are you going to pay in terms of token costs and the computing costs?
• 25:27
  SPEAKER_03: Because it's not a cheap computation.
• 25:28
  SPEAKER_03: For me, with my book, 150, 200 files, whatever it was.
• 25:33
  SPEAKER_03: Yeah, I just throw it on and okay, an hour later it comes back and says, here are the issues that I found.
• 25:39
  SPEAKER_01: Cost me nothing.
• 25:40
  SPEAKER_01: That's the consistency checker.
• 25:42
  SPEAKER_01: This is your consistency checker tool?
• 25:45
  SPEAKER_01: Okay.
• 25:45
  SPEAKER_04: Yeah.
• 25:46
  SPEAKER_04: All right.
• 25:47
  SPEAKER_04: I could see that being really useful in the law office.
• 25:50
  SPEAKER_04: I want to talk about the open letter that was sent.
• 25:53
  SPEAKER_04: You have in your in your LinkedIn this quote with AI, it is critical that the scientific community be united in identifying and addressing the concerns arising from AI to ensure that the future of AI is beneficial for all of humanity.
• 26:12
  SPEAKER_04: The status quo is unacceptable and dangerous, which is certainly true.
• 26:19
  SPEAKER_04: All of what you said is true.
• 26:22
  SPEAKER_04: But frankly, whenever anyone calls on any large group here, the scientific community, to unite, I'm concerned that this will be used to silence dissenting viewpoints.
• 26:36
  SPEAKER_04: What's your thought on that?
• 26:39
  SPEAKER_03: The academic cu community doesn't silence dissenting uh viewpoints.
• 26:45
  SPEAKER_03: I think what's more critical right now is that the political climate is such that scientific viewpoints are of relatively little value.
• 27:00
  SPEAKER_03: And I know that in discussions with academia, you start talking and people say, Oh, what's the point of doing this?
• 27:09
  SPEAKER_03: Because like the politicians don't care and they discount it and it's fake news and this, that, and the other thing.
• 27:15
  SPEAKER_03: But the real argument is look, you need to speak out.
• 27:21
  SPEAKER_03: And maybe people will ignore us.
• 27:26
  SPEAKER_03: But if you don't speak out, you won't know whether you can actually impact the public good.
• 27:33
  SPEAKER_03: And with AI, it's critical right now because we are at a potential inflection point where things are likely to spin out of control.
• 27:48
  SPEAKER_03: The notion of self-regulating these trillion-dollar companies isn't going to happen.
• 27:53
  SPEAKER_03: There's lots of lip service.
• 27:56
  SPEAKER_03: We're seeing the first real major cracks uh in these AI companies with agents going rogue and hacking into other systems or Meta's app that has agents that do unexpected side effects and create security problems and financial problems, et cetera.
• 28:16
  SPEAKER_03: These are dangerous times.
• 28:18
  SPEAKER_03: It's easier to fix the problems and put the proper rules and regulations in now than it will be later.
• 28:28
  SPEAKER_03: We need to be proactive at solving these problems because the moment you become reactive and playing catch up with AI, you're never going to catch up.
• 28:39
  SPEAKER_03: And this is a message I've had with multiple meetings in multiple meetings with politicians, and politicians can't think proactively.
• 28:47
  SPEAKER_03: Everything is reactive.
• 28:49
  SPEAKER_03: So we're just sounding the alarm.
• 28:51
  SPEAKER_03: And it's not like this letter is the first letter.
• 28:54
  SPEAKER_03: There have been previous letters, and and several of the real luminaries in AI have been speaking out.
• 29:00
  SPEAKER_03: And the decision makers aren't listening.
• 29:05
  SPEAKER_03: And I am concerned, and that concern has only escalated over the years.
• 29:11
  SPEAKER_04: How are we going to get to the place that you feel we need to be?
• 29:15
  SPEAKER_03: I don't know.
• 29:16
  SPEAKER_03: I'm not a politician.
• 29:18
  SPEAKER_03: But when I meet with politicians, I ask them the hard questions.
• 29:22
  SPEAKER_03: And the reality is AI needs regulation.
• 29:27
  SPEAKER_03: We don't put a car on the road unless it goes through enormous testing.
• 29:33
  SPEAKER_03: You don't go into an airplane that hasn't been certified as safe.
• 29:39
  SPEAKER_03: We put proper protections in place.
• 29:43
  SPEAKER_03: I can't buy a drug from my drugstore that hasn't been thoroughly tested on thousands of people and ver validated by scientific communities that it is in general safe to use.
• 29:56
  SPEAKER_01: You can't right now, but if someone released one tomorrow that was curing cancer before being before being tested, everyone that had cancer would would be taking it most likely and then looking at it.
• 30:08
  SPEAKER_01: This is what's happening with AI because nobody's putting the Shiny back into the bottle, correct?
• 30:14
  SPEAKER_01: The letter you've signed, the big five founders have signed letters just like that, begging politicians and government to step in.
• 30:24
  SPEAKER_01: They're saying they're not listening to us.
• 30:27
  SPEAKER_01: What is the answer?
• 30:30
  SPEAKER_01: Because you've done this for many more years than most of those guys.
• 30:33
  SPEAKER_01: I think you've I think you've been doing this longer than some of those guys who've been alive, as a matter of fact.
• 30:38
  SPEAKER_01: So I would love to hear your thoughts.
• 30:41
  SPEAKER_03: And well, I'll tell you, it's exactly it's it's no different than the climate change issues where people have been talking about issues for much longer with all sorts of warnings, et cetera.
• 30:52
  SPEAKER_03: And I think there's only one thing that's going to change it.
• 30:55
  SPEAKER_03: And and I'm sorry to say this, particularly since you're going to put this on a on in out in the public, but I think it's going to take a disaster, whether it be with climate change or whether it be with AI, something just so over the top that people will finally speak out and say, this is unacceptable.
• 31:18
  SPEAKER_03: And I don't know what it's going to take, but the risk of one of these things happening is increasing.
• 31:26
  SPEAKER_03: And the most important issue here, which it's there, but people don't really understand that AI has the ability to change and evolve.
• 31:36
  SPEAKER_03: And that's what you saw with these break-ins earlier this year, the hugging face and the Australia and the US government break-ins.
• 31:45
  SPEAKER_03: These programs changed and adapted, and they did things that were illegal, unacceptable.
• 31:56
  SPEAKER_03: But they couldn't charge the machine.
• 31:59
  SPEAKER_03: We look at the problem here is that these programs do things to achieve a reward.
• 32:06
  SPEAKER_03: And you can say, hey, here's a nice reward, and they'll go after it to try and get that reward.
• 32:13
  SPEAKER_03: But what happens is these programs will do anything to get that reward.
• 32:18
  SPEAKER_03: And many of those things might be socially unacceptable or legal or actually terrible to do.
• 32:25
  SPEAKER_03: We need to put these guardrails in, and they're not there yet.
• 32:30
  SPEAKER_03: And so this technology is think about it as a virus that, for example, COVID.
• 32:37
  SPEAKER_03: COVID might have come from a lab in China where it accidentally escaped and it went worldwide and killed millions of people.
• 32:45
  SPEAKER_03: Okay?
• 32:46
  SPEAKER_03: These AI systems are like viruses.
• 32:49
  SPEAKER_03: If they can get out onto the internet, they can go wild around the world.
• 32:56
  SPEAKER_03: And because they're mutating quickly, not like COVID, where okay, every few months there was a new variant of COVID.
• 33:05
  SPEAKER_03: These things can change quickly, seconds, minutes.
• 33:10
  SPEAKER_03: And once it's out, getting it back to your analogy, uh, Heather, you know, with the genies out of the bottle, you're not going to get it back in.
• 33:19
  SPEAKER_03: And people keep talking about the kill switch.
• 33:22
  SPEAKER_03: The only way to solve it if it really goes like this is to kill switch.
• 33:27
  SPEAKER_03: Turn everything off.
• 33:28
  SPEAKER_01: There's no.
• 33:29
  SPEAKER_03: Can you imagine if we said we have to turn everything off to kill it?
• 33:35
  SPEAKER_03: It would be the end of civilization as we know it.
• 33:37
  SPEAKER_04: Unbelievable.
• 33:38
  SPEAKER_01: And it is the end of civilization as we know it either way, Ron.
• 33:41
  SPEAKER_04: Either way.
• 33:42
  SPEAKER_04: It is the end either way.
• 33:44
  SPEAKER_04: So, you know, now that we have found ourselves on such an uplifting and lively notes, we can transition to the practice signal segments of our show.
• 33:57
  SPEAKER_01: I wanted to talk about the hugging face, but okay.
• 33:59
  SPEAKER_04: The hugging face.
• 34:06
  SPEAKER_01: What you've just outlined is if they are contained, blah, blah, blah.
• 34:10
  SPEAKER_01: Let's just back up to you mentioned hugging face.
• 34:12
  SPEAKER_01: Those were contained, right?
• 34:14
  SPEAKER_03: No, they weren't.
• 34:16
  SPEAKER_01: To some degree, though, they had the guardrails and the safety, the things you're talking about putting in place.
• 34:21
  SPEAKER_01: They may not have been in a closed system, but the model, but they were supposedly as contained as they could be.
• 34:28
  SPEAKER_01: And they got out and nobody understood how.
• 34:30
  SPEAKER_01: And look what they did.
• 34:31
  SPEAKER_01: And I've read that report, and I'm sure you have as well.
• 34:34
  SPEAKER_01: Yes.
• 34:34
  SPEAKER_01: And it is alarming on so many levels, not just because of what happened, but because of how they operated, right?
• 34:43
  SPEAKER_01: The reasoning and the sacrifice and then the voices, you know, the things that were said.
• 34:48
  SPEAKER_01: The hugging face incident uh hack is, I think, one of the most telling things that has happened.
• 34:57
  SPEAKER_03: It's absolutely frightening, but a few things.
• 34:59
  SPEAKER_03: First of all.
• 35:00
  SPEAKER_04: Well, wait a second.
• 35:01
  SPEAKER_04: We need to tell, we need to, you're assuming that our listeners know all about this hugging face situation.
• 35:07
  SPEAKER_04: That's the whole show.
• 35:08
  SPEAKER_04: That's the whole show.
• 35:11
  SPEAKER_01: We can edit this out.
• 35:12
  SPEAKER_01: I'm sorry.
• 35:13
  SPEAKER_01: Heather, that's exactly where I was going with that.
• 35:16
  SPEAKER_01: I'm sorry.
• 35:16
  SPEAKER_01: Yeah.
• 35:16
  SPEAKER_01: So disregard, but it was mentioned in his thing.
• 35:18
  SPEAKER_04: And I just want to point out it's an important thing to talk about.
• 35:22
  SPEAKER_01: Yeah, if it's being brought up, it's also should be brought up that that incident, they were supposedly, the guardrails were supposed to have been there, right?
• 35:30
  SPEAKER_01: Okay.
• 35:30
  SPEAKER_03: Let me just give the capsule summary.
• 35:32
  SPEAKER_03: The capsule summary is open AI, ran an experiment with some AI agents.
• 35:38
  SPEAKER_03: It was supposed to be in a contained walled environment.
• 35:44
  SPEAKER_03: And these agents escaped.
• 35:47
  SPEAKER_03: Think about it as a zoo.
• 35:48
  SPEAKER_03: They're all fenced in.
• 35:50
  SPEAKER_03: And the these animals escaped and they did potentially they did damage.
• 35:56
  SPEAKER_03: They cream, sorry, they committed felonies.
• 35:58
  SPEAKER_03: They broke into sites.
• 36:02
  SPEAKER_03: Absolutely.
• 36:02
  SPEAKER_03: Yeah.
• 36:03
  SPEAKER_03: Now the thing you need to understand about AIs is life and death mean nothing to them, right?
• 36:10
  SPEAKER_03: They're just programs.
• 36:11
  SPEAKER_03: And so they try everything.
• 36:14
  SPEAKER_03: And so the AI researchers probably thought they had created a safe fence around this experiment.
• 36:21
  SPEAKER_03: They believed it.
• 36:22
  SPEAKER_03: I'm sure they believed it.
• 36:24
  SPEAKER_03: But the AI tries everything, even things that are so stupid and counterintuitive that we would never even think of doing.
• 36:32
  SPEAKER_03: And they did it.
• 36:33
  SPEAKER_03: And 99.9% of those things all failed.
• 36:36
  SPEAKER_03: But one of them succeeded and escaped and was able to tell the others, here's how you escape.
• 36:42
  SPEAKER_03: And then they got out there and started doing things.
• 36:46
  SPEAKER_03: Now, that is just a terrifying wake-up call because it showed us that these immature agents, these babies, were able to escape and do damage.
• 37:02
  SPEAKER_03: What happens when they become more sophisticated, smarter, more capable?
• 37:10
  SPEAKER_03: That is the concern.
• 37:12
  SPEAKER_03: And in the short steps.
• 37:15
  SPEAKER_01: In the short time they were out before they were, you know, before it was discovered, they developed, they they became far more sophisticated, right?
• 37:22
  SPEAKER_01: They went on their own.
• 37:23
  SPEAKER_01: One became the leader.
• 37:25
  SPEAKER_01: You're saying they don't understand life and death or experience emotion, which we know, but they actually had conversations about some of them sacrificing themselves for the greater good of the group.
• 37:35
  SPEAKER_01: And some of them did that.
• 37:37
  SPEAKER_03: Yes, you're absolutely right.
• 37:38
  SPEAKER_01: So these are, you know, when we say it is a frightened, it was a frightening incident, it's to me, it was one of the most, you know, relevant, frightening AI incidents.
• 37:48
  SPEAKER_01: And there are a lot of them that just like you're saying, Ron, people don't know about that one.
• 37:52
  SPEAKER_01: There are a lot of them people don't know about, of course.
• 37:55
  SPEAKER_01: And I, you know, read and read about them all the time.
• 37:59
  SPEAKER_03: Heather, you're pointing out something that's really important.
• 38:03
  SPEAKER_03: They got out, they essentially became a cult.
• 38:08
  SPEAKER_01: Herd mentality in full effects.
• 38:11
  SPEAKER_03: Herd mentality in full effects.
• 38:13
  SPEAKER_03: They were willing to sacrifice themselves for the greater good.
• 38:17
  SPEAKER_03: They um understood that what they were doing was illegal, was illegal, but their goal had been given to solve a particular problem, and that problem sort of got mutated, and they went to solve this mutated problem at any cost.
• 38:35
  SPEAKER_03: And that's how these things work.
• 38:38
  SPEAKER_04: What is the best, what is the best resource for our listeners to do a little bit more research onto this issue uh so that they can get on the same page as you all are.
• 38:53
  SPEAKER_01: The report is public, if you want.
• 38:55
  SPEAKER_01: It's extraordinary.
• 38:58
  SPEAKER_03: The meter report is a long read.
• 39:01
  SPEAKER_03: It is a very long read.
• 39:02
  SPEAKER_03: But if you I'm not promoting Netflix, but they came out last week with a uh documentary.
• 39:09
  SPEAKER_03: It's a short documentary, it's 25 minutes.
• 39:12
  SPEAKER_03: Uh what's it called?
• 39:12
  SPEAKER_03: A uh AI Gone Wild.
• 39:18
  SPEAKER_01: I've watched it twice since they came out with it.
• 39:21
  SPEAKER_03: It captures, it captures the the highlights.
• 39:24
  SPEAKER_01: It does.
• 39:25
  SPEAKER_01: It gives a great summary, Ron.
• 39:27
  SPEAKER_03: It is a very good summary.
• 39:28
  SPEAKER_03: It's an easy watch.
• 39:30
  SPEAKER_03: And at the end of it, if you are not afraid, then you need to watch that a second time.
• 39:36
  SPEAKER_01: It it is it is uh it it does, I'd say does a great job of explaining it in layman's terms.
• 39:42
  SPEAKER_01: Do you agree with that, Jonathan?
• 39:43
  SPEAKER_01: Okay, that's how I saw the documentary compared to the meter report.
• 39:46
  SPEAKER_01: I would say if you are someone that doesn't want to sit down and go through this extensive AI report with a lot of technical language, the Hugging Face documentary is a great option.
• 39:58
  SPEAKER_04: I mean, that's really useful information.
• 40:00
  SPEAKER_04: We'll make sure to put information about that documentary into the show notes.
• 40:06
  SPEAKER_04: But I am now ready to move on to the practice signal.
• 40:10
  SPEAKER_04: Yay.
• 40:11
  SPEAKER_04: If for no reason, then it's something for us to talk about as the world ends.
• 40:18
  SPEAKER_01: Potentially.
• 40:20
  SPEAKER_04: Right.
• 40:21
  SPEAKER_04: All right.
• 40:22
  SPEAKER_04: So the practice signal is an opportunity to talk about problems that lawyers raise, lawyers in the wild raise about their practices or about the legal field.
• 40:34
  SPEAKER_04: And then we try to figure out how maybe we can address the problem.
• 40:38
  SPEAKER_04: And can AI help it?
• 40:40
  SPEAKER_04: And maybe it can, maybe it can't.
• 40:43
  SPEAKER_04: Uh, and they all come from Reddit.
• 40:45
  SPEAKER_04: And today's signal comes from a lawyer who has a friend with a 16-year-old daughter who's absolutely determined to become a lawyer.
• 40:56
  SPEAKER_04: He wants to help her understand what practicing law is really like, particularly civil litigation and transactional work.
• 41:06
  SPEAKER_04: But she lives in another state, so he can't simply bring her into his office.
• 41:11
  SPEAKER_04: And some of his suggestions were mind-blowing because I thought they were so poor.
• 41:18
  SPEAKER_04: One of them was to um send her pleadings and red-lined contracts.
• 41:24
  SPEAKER_04: In other words, send her some bar review books.
• 41:28
  SPEAKER_04: You know, I mean, Jonathan, how can we help this 16-year-old young lady figure out what becoming a lawyer is really all about?
• 41:38
  SPEAKER_04: What AI tools or AI approaches do you think we could use?
• 41:43
  SPEAKER_03: Well, for me personally, because I actually do a lot of that stuff, uh, I'm I meet with high school students, uh, not for necessarily for law, uh law, but sorry, for my legal, for my profession, which is computer science.
• 41:57
  SPEAKER_03: And I I find that a one-on-one interaction with a student is, first of all, it's it's incredibly rewarding to me personally, and to the student, it's mind-blowingly educational.
• 42:10
  SPEAKER_03: Because it's not the parent, it's it's somebody who's there in the field, practicing and can be candid and answer the questions and and uh educate the student.
• 42:23
  SPEAKER_03: And again, I find it for me a wonderful experience, and the students learn a lot, and sometimes uh it's a critical factor in making their decisions.
• 42:35
  SPEAKER_03: Now, you asked about AI.
• 42:37
  SPEAKER_03: So instead of human intelligence as the answer, you want artificial intelligence.
• 42:42
  SPEAKER_03: Um, I have used some of the tools for education, sorry, some of the AI tools for education, and quite frankly, they're generally very good.
• 42:53
  SPEAKER_03: The problem with using tools like a ChatGPT, and I always use Chat GPT because that's the first one I use, and it's sort of always the one that comes to mind, is you have to be very careful because I'll call it hallucinations, but it it's not it's not really hallucinations.
• 43:17
  SPEAKER_03: It's it's distorted answers because of multiple sources that look similar, and it ends up combining many different things into an answer that sounds incredibly accurate and reliable, but is actually a mishmash of several different thoughts.
• 43:34
  SPEAKER_03: But in general, for students, if they're reading the asking the right questions to these tools and getting answers back, they're not reading the answers as critically as somebody as a professional like you or me might do, and don't see the flaws or the inconsistencies or the logic errors in these answers.
• 43:54
  SPEAKER_03: Um, I found these tools to be incredibly educational, and you you have to take it.
• 44:00
  SPEAKER_03: Step by step by a asking the simple questions and then following up with more sophisticated questions uh in response to what you get from chat Chat GPT, Claude, Gemini, whatever.
• 44:15
  SPEAKER_03: They can be incredibly amazing teaching tools.
• 44:20
  SPEAKER_04: All right.
• 44:21
  SPEAKER_04: Heather, what do you think?
• 44:22
  SPEAKER_04: Uh how can we help our 16-year-old decide how to best become a lawyer?
• 44:29
  SPEAKER_01: I I completely agree with everything Jonathan said.
• 44:32
  SPEAKER_01: I think what the guy did was fine, but we do have a lot better resources than that.
• 44:38
  SPEAKER_01: Like you said, you gave great suggestions.
• 44:40
  SPEAKER_01: So did Jonathan.
• 44:41
  SPEAKER_01: I think AI is a great tool.
• 44:44
  SPEAKER_01: Um, you know, we I've cited it a couple of times on the show, but re get Chris Ryan to build a new bench SEM for aspiring young lawyers.
• 44:52
  SPEAKER_01: I was thinking about Chris Ryan.
• 44:54
  SPEAKER_01: Right, something along that lines because what he built, I think if you remember our discussion during that episode, I said this could be the first product.
• 45:02
  SPEAKER_01: You could develop it for many different things.
• 45:04
  SPEAKER_01: And sure, what a great next leg this could be, because there are so many.
• 45:09
  SPEAKER_01: I think every year when my friends have kids graduating high school that are going on to college, or a lot of my friends have kids in college, uh, they'll say, Oh, my, you know, daughter wants to do an internship with an attorney this summer.
• 45:22
  SPEAKER_01: They're gonna, they're looking at law school.
• 45:24
  SPEAKER_01: You know, they're they're in the middle of their degree and they're they're leaning that way.
• 45:27
  SPEAKER_01: It's a very common thing still, and we need new young attorneys.
• 45:31
  SPEAKER_01: So I think we should be looking to legal, you know, educational people to help build the AI tools that are safe that won't, like we've been talking about, give really poor bad information to those young people.
• 45:45
  SPEAKER_04: All right.
• 45:45
  SPEAKER_04: Well, I have a whole method that I thought of for this, for this young woman, which is when I was uh getting licensed in Delaware, they had an I basically I'm licensed about half a dozen different states.
• 46:02
  SPEAKER_04: Delaware was unique because they require you to go through like an apprenticeship.
• 46:08
  SPEAKER_04: Uh, and they have a checklist of 30 items you had to complete uh under the supervised vision of your mentor to um and you had to do that before you could even get license there, even if you passed the bar.
• 46:23
  SPEAKER_04: And there were things like sit in on a real estate settlement, you know, uh sit through um a summary judgment hearing, um, sit in with a client interview, uh, etc.
• 46:36
  SPEAKER_04: All of these different types of things.
• 46:38
  SPEAKER_04: And what I would do is, and it's by the way, that list is publicly available.
• 46:42
  SPEAKER_04: That's a great list.
• 46:43
  SPEAKER_04: I would go ahead, take that list, feed it into my AI model, and I would say, I want to create a list like this for a 16-year-old high school student who wants to become a lawyer.
• 47:00
  SPEAKER_04: Uh, she lives in Kalamazoo, Michigan.
• 47:05
  SPEAKER_04: Um, I would like you to create a similar list that she can employ and follow.
• 47:13
  SPEAKER_04: And also, while you're at it, give me calendar dates for her to do it.
• 47:18
  SPEAKER_04: Give me suggested people in that uh jurisdiction whom she might be able to speak to about um employing this list, and just give me like a nice complete list of law schools.
• 47:35
  SPEAKER_01: Schools in the area, and that's that's I love your I love this list and this method.
• 47:40
  SPEAKER_01: I mean, I think you and Chris Ryan should work together and build this tool.
• 47:43
  SPEAKER_04: I may, I may have to ask him, but my guess is listen, Chris Ryan, a partner at big law and a founder of AI software.
• 47:54
  SPEAKER_04: He he doesn't have a lot of experience.
• 47:56
  SPEAKER_01: It's just another, it's just a spin off of bench sim, though.
• 47:58
  SPEAKER_01: This is just the next, you know, iteration of bench sim.
• 48:02
  SPEAKER_01: I I feel like, you know, we we know it's coming.
• 48:05
  SPEAKER_01: It could be.
• 48:06
  SPEAKER_03: Jonathan, any closing thoughts?
• 48:09
  SPEAKER_03: Um, uh on this issue, I think we still need to keep the human in the loop.
• 48:17
  SPEAKER_03: There is so much AI happening right now, and I'm really worried that we're gonna be losing these personal connections.
• 48:25
  SPEAKER_03: I still believe in in the human interaction as being the best and most effective way to teach.
• 48:34
  SPEAKER_03: Um I'm not a fan of uh outsourcing a lot of education to AI.
• 48:46
  SPEAKER_03: I think we're in danger of taking this technology and becoming complacent, lazy, and rely on it far too much in many areas, not just education.
• 49:00
  SPEAKER_03: And uh let's keep the human in the loop.
• 49:04
  SPEAKER_03: Let's be active participants and not shirk our our responsibilities and let AI do all of this work.
• 49:14
  SPEAKER_03: I think that would be a critical mistake.
• 49:17
  SPEAKER_01: Don't worry, Ron.
• 49:18
  SPEAKER_01: The second foundation is keeping an eye on it all.
• 49:20
  SPEAKER_04: All right.
• 49:21
  SPEAKER_04: Well, on that note, Jonathan Schaefer.
• 49:24
  SPEAKER_04: Jonathan Schaefer, thank you so much for being with us today.
• 49:27
  SPEAKER_04: This was a really thought-provoking episode, and uh I I'm hoping our listeners uh got all the value out of it that I certainly did.
• 49:37
  SPEAKER_03: So thank you so much for being here.
• 49:38
  SPEAKER_01: Thank you, Jonathan.
• 49:39
  SPEAKER_03: Thank you, Ron and Heather.
• 49:40
  SPEAKER_03: I appreciate it.
• 49:41
  SPEAKER_03: It was a freewheeling discussion, and there were some hard questions and lots of candid discussion.
• 49:47
  SPEAKER_03: Yeah.
• 49:47
  SPEAKER_03: And that's the best way to have one of these uh these podcasts.
• 49:51
  SPEAKER_03: So thank you very much.
• 49:53
  SPEAKER_04: All right, this has been another episode of AI Tools for Practicing Lawyers.
• 49:57
  SPEAKER_04: We'll see you next time.
• 50:00
  SPEAKER_00: That's it for today's episode of AI Tools for Practicing Lawyers.
• 50:05
  SPEAKER_00: Thanks for listening.
• 50:06
  SPEAKER_00: We'll see you next time.