Latest / Newcomer Pod / The AI Personal Stack (with Joshua Browder)
Transcript
- 0:10Hey, it's Eric Newcomer. Welcome to Cerebral Valley
- 0:14newcomer. Back.
- 0:15With me are Max Child and James Wilsterman, the Co founders of
- 0:19Volley. Hey guys.
- 0:21Welcome back. Hey, happy to be here.
- 0:24Hey, Eric. Hey.
- 0:26All right. This week we are talking about
- 0:28what we're calling the AI Personal Stack.
- 0:32Or who? How do we actually use AI in our
- 0:36real lives? I've got in the second-half of
- 0:40this episode, Joshua Browder, the CEO of Do Not Pay, who got
- 0:45in trouble trying to use AI to act as a lawyer.
- 0:49He got his start trying to get out of parking tickets that he
- 0:54increasingly has automated, and now he's a ChatGPT addict,
- 0:58building a company around using foundation models and language
- 1:03models to fight back against companies like Comcast and your
- 1:08mega corporate gym. I think Equinox is a big target
- 1:12of his just helping people get out of the bill, their bills.
- 1:15So that's a fun conversation from someone who's really trying
- 1:19to suck the marrow out of AI and get everything he can out of it.
- 1:24So, yeah, excited to talk to you guys about how how we're using
- 1:29AI, how we're seeing people actually use it.
- 1:33Because, you know, in Silicon Valley, people's expectations,
- 1:38hopes and dreams can get ahead of them.
- 1:40We can all live in the world of intuitions as we did in the last
- 1:45episode, where it's like, what could be possible?
- 1:47Where could we take this? You know, it's an optimistic
- 1:50sort of place and this episode is going to try and ground us
- 1:54in. How are we using it?
- 1:56Are people using it? So, So that's the question,
- 2:01which of us, who thinks they use AI the most?
- 2:05I honestly would not put myself super, super high on this list.
- 2:09I guess probably James. I'm going to go, James.
- 2:12Yeah. OK.
- 2:12All right. There we go.
- 2:13I have the same guess here, James, given that we believe
- 2:17that you probably use it the most and I think you had an
- 2:20exciting AI driven experience today.
- 2:24How are you using AI in your life?
- 2:27That is very true. I commuted today in a Waymo,
- 2:31which was my first time in a Waymo.
- 2:33I've taken a few cruises around San Francisco.
- 2:36I think that that those self driving car experiences really
- 2:40ranked at the maybe the pinnacle to me of excitement of using
- 2:45AIII would say GPT chat. GPT is, you know, very.
- 2:52Very useful and I I use it all a lot, but I feel like the
- 2:55visceral excitement of riding in a self driving vehicle and just
- 2:59taking that to work is something that I have haven't experienced
- 3:03elsewhere with. Could you tell the difference
- 3:04between Cruise and Waymo? Was there a spiritual or is it
- 3:08just how the interior of the car feels or?
- 3:11I I guess I do think of it as I don't want to like over learn,
- 3:15you know from a few anecdotes of riding in in these vehicles,
- 3:21but. I do feel like there is a bit of
- 3:24AI feel like the Waymo felt a little bit more natural, a
- 3:26little bit less cautious, but not in a scary way.
- 3:30I feel like Cruise is very cautious, driver cautious, yeah.
- 3:34Yeah, and neither. No humans beside you, right?
- 3:39Like no driver. Correct.
- 3:41And and interesting in the Waymo is I was able to sit in the
- 3:44front seat. You're not able to do that in a
- 3:46cruise. And that was it.
- 3:48That alone was a pretty amazing experience because.
- 3:51I was, I just like could turn left and there was nobody there.
- 3:55And you know, I could see the whole road and see see exactly
- 3:59what was happening. So I feel like I enjoyed the way
- 4:03MO experienced a lot today. And I I think being able to ride
- 4:07up front, having it feel like more human in some sense, that
- 4:12it just kind of was relaxing, that I felt like I I wasn't.
- 4:17I wasn't like wasting time by taking a Waymo to work because
- 4:23you know it, it drove pretty naturally.
- 4:25And yeah, it was an amazing experience.
- 4:27And you know, the Waymo CEO, Co CEO, Dimitri is speaking at
- 4:32Cerebral Valley. I think we will have announced
- 4:34that by the time this podcast comes out, Yes.
- 4:38But so very excited to hear more.
- 4:40Very. Excited as well other other
- 4:42uses. Of it, Yeah.
- 4:45So. Definitely.
- 4:47I've already talked on this podcast about using ChatGPT for
- 4:49just creative kind of brainstorming and exploration.
- 4:53I think that it's very effective at creating kind of internal
- 4:56documentation or early drafts of it or getting you started on
- 5:01that at work. And then I also think that the
- 5:05code side is like really, really important here.
- 5:08A lot of Volley engineers are using ChatGPT in their
- 5:12day-to-day experience coding at Volley and.
- 5:15What that looks like is pretty interesting because it's
- 5:18primarily kind of bouncing between your IDE and IDE, your
- 5:26independent development environment, integrated
- 5:30integrated integrated development environment.
- 5:32Where you code basically and bouncing between that and sort
- 5:37of ChatGPT. Where you might ask for an
- 5:40architectural advice, or even paste code in, or ask for some
- 5:44code in. It used to be you Google Stack
- 5:45Overflow, blah blah blah and they'd copy and paste it there.
- 5:48Exactly. And it's it's a really similar
- 5:51experience to when I was learning to code.
- 5:53And I think what is interesting is a lot of even more junior
- 5:56developers are the the most eager to kind of explore here
- 6:00and just use ChatGPT because they want to learn quickly and
- 6:03you know they don't, they were going to Google it anyway from
- 6:06Stack Overflow so. I think that is a huge change in
- 6:10how especially junior software developers are are working
- 6:13today. And obviously the promise there
- 6:15is like integrated code environments where.
- 6:18Yeah, and there is a start up called Cursor that has a
- 6:23integrated IDE and I know it's paying for that.
- 6:27No, we're not doing it. We're not using any of anything
- 6:29like that other than GitHub Copilot, which is I think you
- 6:34know I use GitHub Copilot. It essentially is auto complete.
- 6:38For code, and I would say it's useful but not revolutionary,
- 6:44not game changing. It maybe improves your kind of
- 6:47speed by, you know, 5% or something, anything else.
- 6:51I actually will give you another one which I think is really
- 6:54interesting. I am a new father.
- 6:56I had a newborn born about a month ago, and congratulations.
- 7:02Thank you and. I find chat chi PT to be really
- 7:06incredible as a parenting assistant.
- 7:09Just being able to ask questions like when should I expect my
- 7:14baby to be able to see certain distances right.
- 7:18And I could Google that and and I'd get different sources and I
- 7:23you know it would be you know something I could research but I
- 7:27basically don't need like a precise answer and and Chat chi
- 7:30PT is able to give me hey here's how the here's how this might
- 7:32look over the next. Month, two months, six months,
- 7:35right? That's just really convenient to
- 7:38be able to ask a question in open-ended natural language like
- 7:41that. All right, Max, what about you?
- 7:44How are you using? I mean a lot less than James, so
- 7:47I'll move through this in a hurry I would say.
- 7:49One I occasionally have it right, like memos or long form
- 7:53emails for Volley. But again, I would say only in
- 7:56the case where I'm OK if it's like B plus quality output
- 7:59'cause if it really has to be an A or an A plus, I feel like I
- 8:02have to do it myself. And then?
- 8:04I dictate like texts and emails and messages like quite a lot
- 8:07like I would say an old man. Yeah, I'm like an old man.
- 8:11I mean, I'm, you know, I have a kid too.
- 8:12I'm, I'm in my 30s, so that's old enough.
- 8:15Yeah. I would say at home I dictate a
- 8:17lot of texts and messages into my phone and I now think that
- 8:20like Apples native dictation is like.
- 8:23You know, 98% accurate or whatever you want to call it and
- 8:26and if I want to dictate a, how do you actually do it?
- 8:28You just hit the little microphone button in the bottom
- 8:30right of the keyboard and then you just start talking right
- 8:32away and it turns into text. Turns into text.
- 8:35It's pretty fast and it's pretty good.
- 8:37So I do that a lot. Other than yet that.
- 8:40I mean, yeah, I I sometimes use like some of those image
- 8:44generators for for internal, you know, slide decks, but that's
- 8:47usually almost as like a joke because they're sometimes
- 8:49hilarious. So.
- 8:51I mean that's kind of it. I am playing with some of these
- 8:55newer apps that are based around image generation.
- 8:57Like I think we talked about Can of Soup, which is like a very
- 9:00hot YC company that just came out that is basically like
- 9:04Instagram for image generated content.
- 9:07And so every day they send you a push notification that says
- 9:10like. Max riding a giant chicken, like
- 9:13in a cowboy hat or something, and then you click on it and it
- 9:16generates a photo of you, you know, writing, writing a giant
- 9:18chicken on a cowboy hat or wearing a cowboy hat.
- 9:21And you know, do you use that everyday?
- 9:23Did you use? That I clicked on that.
- 9:25I don't think I'm like a daily user, but I've used that a
- 9:28couple times a week because they're pretty good at sending
- 9:30you prompts that just sound funny, like.
- 9:33And so I click on it like maybe one to two out of seven,
- 9:35basically and. I don't really use it for the
- 9:38Instagram functionality of like looking at other people's
- 9:41because I don't really care like what other people look like on
- 9:43top of giant chicken, like like everyone.
- 9:45I'm very a selfish person and so yeah, I like.
- 9:47But that use case of just sending you a prompt everyday
- 9:51and then just you click on it and a push notification opens
- 9:54right out to generating something from that prompt is a
- 9:56fairly compelling like use case. I don't know how you build a
- 10:01social network out of it, but that part has been working on me
- 10:03a little bit, Eric. Yeah, Eric, Eric.
- 10:09So Eric how do you use AI in your in your daily life?
- 10:14I mean I'm probably the most disillusioned with Chad GB TI
- 10:18mean credit credit to the AI. Then in the beginning when when
- 10:23image generation was new I used some of it to illustrate my
- 10:26newsletter which is not you know that's significant it it would
- 10:30probably cost a lot. You know, I I get access to
- 10:32Getty Images through Sub Stack, so I don't have to agonize about
- 10:37the price of images. But you know, there's meaningful
- 10:39value to that. And like it's a big editorial
- 10:42decision to decide to put it out.
- 10:43Though I do think there was a period where like where the
- 10:46aesthetic of AI was cool and sort of even if it was bad, it
- 10:51was showing that you were sort of you were doing sort of AI
- 10:55stuff. And eventually obviously the
- 10:58images need to stand on their own and be great.
- 11:01And I partially, I just need to be better at prompting and I
- 11:04think I'm just not like expert level and it's so it's not worth
- 11:07the energy to use it. I've found Chachi BT to be very
- 11:11successful in proofreading. The most impressive experience
- 11:16being that inputted a story and it had it told me that a million
- 11:22was supposed to be a billion which was like, OK, wow, that's
- 11:25like that was great. That would have been like,
- 11:28that's hard to catch. I think it's checked.
- 11:31It'll catch like internal misspelling errors, you know, in
- 11:35proper nouns, you know. Obviously getting names right is
- 11:37always important. It's good at yeah, it'll, it
- 11:42catches like double word typos pretty well.
- 11:45So I think it's always useful putting that in.
- 11:49It can be good about directing you towards active voice, but I
- 11:54feel like as soon as you start asking for subjective advice it
- 11:59gets pretty bad. And like, I don't, you know?
- 12:02To clarify is that is that 3.5 that you're using?
- 12:05No, no no, not four, only four. I find 3.5 terrible like I I so.
- 12:09I thought you said you gave up on paying for it.
- 12:12I was paying and then I gave up, so I just returned to relying on
- 12:16humans to proofread. But now because we're recording
- 12:19it, so I did give it up like partially just frustration that
- 12:23like I perceived it to be getting worse.
- 12:26Though I'm not sure what I actually think on that front.
- 12:31And I I I tried using web plugins and I actually found
- 12:34them like of poor experience so. It's very slow.
- 12:38Like did you ever try the web browser plug in they they they
- 12:41just added their own like so originally they had an outside
- 12:44one and I think now they have their own.
- 12:47It's still. It feels like it relies on like
- 12:49one page and then. I agree it's it's not very
- 12:53effective because it's essentially.
- 12:55Going to Bing and Googling whatever you said Bing then.
- 12:59Please, I think. Sorry, that's pretty funny.
- 13:03Oh my God. Poor Microsoft.
- 13:05Poor poor Microsoft. You go to Bing.
- 13:07Bing. Yeah, binging your prompt and
- 13:10then essentially just clicking on maybe one or two links and
- 13:17somehow using that. LLM That text in the LLM process
- 13:21that that just doesn't seem very productive right now.
- 13:24I don't. I hope they can get better at
- 13:25that. I mean, I I've said, you know,
- 13:29it changed my vows. It made like a line and active
- 13:32voice, which was minor but sort of funny that they will forever
- 13:35have been impacted by Chachi BT I I generally find it in writing
- 13:40to be like a siren song in that it feels so easy and when I'm
- 13:45feeling lazy it's like this will generate text.
- 13:48But I honestly think it's like a bad starting point in that it's
- 13:53so off. Like I think I created a job
- 13:55posting with Chachi PT and we haven't heard any.
- 13:59We haven't heard that. I don't know.
- 14:00I blame Chachi. Somebody just told me I need to,
- 14:03like, chill it out. It's a little, you know, like,
- 14:06so I'm going to have a human rework that.
- 14:09I mean, I do, but it's good at, like, things that appear, which
- 14:12is why I'm calling a siren song. Like, it appears good.
- 14:15Like, it's like, oh, that's what a job posting looks like.
- 14:18But it doesn't always feel like it's sort of thought through on
- 14:22every piece of it. Have you tried pasting in an
- 14:26article of yours, like a draft and asking Chachi PT?
- 14:31Hey, is there anything I'm missing?
- 14:32What questions will the reader ask?
- 14:35That is, that's what I mean. I, I, you know even for this
- 14:39show I put in and it has been helpful prepping for the show.
- 14:43Like I put in our outlines and then said blah, blah, blah.
- 14:47Like. I mean it was helpful sort of
- 14:50filling in timelines and coming up with like papers that I think
- 14:53we referenced in the first episode.
- 14:55But I I don't think it's really achieved the mirror my style
- 14:59sort of thing and give me it. I feel like it doesn't adhere to
- 15:03my style enough which is really frustrating given how much of my
- 15:07writing is online that it's I'm like Eric Newcomer, a public
- 15:10figure who's written your XYBC plays like it doesn't really
- 15:15learn that much off of it. And you know, even when I give
- 15:18it a lot of text to prompt, I don't feel like it really gets
- 15:21my style very well. And I do think overall, yeah, I
- 15:26guess I haven't been critical enough so far in this podcast of
- 15:29how much I find the writing and thinking style to be not, not
- 15:35there, like it. I mean, again, I'm open to the
- 15:37idea that there are prompts that have to fix it.
- 15:42Yeah. Anyway, more more negative than
- 15:44I've been on AI, but I just, I feel like the threat to writing,
- 15:50you know, I believe it. If it gets like, you know, the
- 15:53next, if the improvement from to the next version is as great as
- 15:56it was from 3.5. I'm still open to the idea that
- 15:59it becomes a great writer. But I think it's great at
- 16:02writing poems and like more formulaic stuff.
- 16:06Like it's great at rhymes. Like I feel like I've, you know,
- 16:10come up with like funny, like here's the history of me and my
- 16:13wife, like write a poem, but but it, you know, they're, they're
- 16:18sort of gimmicks. Well, going back to the kind of
- 16:21integration into the coding environment idea, do you think,
- 16:27Have you tried any tools like that for writing?
- 16:29I know there's a bunch like writing specific ones.
- 16:32I you know, I know the guys that the guy who's doing Lex, he was
- 16:36sort of an X sub stack or from the newsletter every I need to.
- 16:40I haven't used his yet. Have you guys have either of you
- 16:43used any writing assistant ones? I don't use tome either to be
- 16:46honest, which I want. To try I use tome, yeah.
- 16:48What do you think of it? Not great tome is the tome,
- 16:53didn't you? Which one?
- 16:54Have you interviewed him? You interviewed him last time,
- 16:56right? I mean, I just think that tome
- 16:59is the PowerPoint. To give the generous, yeah, the
- 17:01generous interpretation. So it's basically Keynote or
- 17:04PowerPoint, but with with AI generation as the core creation
- 17:08mechanism. I mean it, it was impressive
- 17:11demo. I think it felt like a very cool
- 17:13demo for again. I mean I think it was more of
- 17:16the middle school, high school level of output, not even the
- 17:19college level of quality. And so if I were like in 6th
- 17:22grade and I needed to make like a book report about like
- 17:25Yosemite National Park or something, I think it's like a
- 17:27killer use case. For the record, that is what the
- 17:30CEO basically said was like the core use case was like homework
- 17:34and and such. So you know, not to give him a
- 17:36hard time, but as like a professional.
- 17:39You know, slide deck creation tool, it was really not there.
- 17:42Like what? Was yeah, what was missing, I
- 17:44guess. I would say to come back to this
- 17:47text generation being the big issue, like the content on the
- 17:50slides was really, really kind of empty and banal, right?
- 17:55It was like like it was the most sort of vague, like tautological
- 18:01bullet points, like in the slide deck and then.
- 18:04And then it was putting in generated artwork, right, Which
- 18:07again I used this like four or five months ago.
- 18:09So it was like pre the last two generations of mid journey or
- 18:12whatever. But the AI generated artwork was
- 18:14like really bad in like March and April.
- 18:17And so you were like well I could never use this in a real
- 18:19presentation. And so at some point it was sort
- 18:21of like the only thing this tool did for me was like make 10
- 18:24slides and like put essentially bullet points and I'm going to
- 18:29have to rewrite on them. And so it was like I could do
- 18:31this by. Copy pasting slides in in a
- 18:33normal presentation, you know framework and that's what I and
- 18:36in some cases having a starting point can be helpful, but in
- 18:39other cases it just sort of constrains your thinking
- 18:43unnecessarily. Yeah, I mean, yeah, I even if it
- 18:48doesn't constrain my thinking, to your point, it's sort of a
- 18:50siren song where I like, I'm like, oh cool, someone else is
- 18:52going to make my slide deck and then they make it slide deck and
- 18:54I'm like, OK, well, I can use about 10% of this, so.
- 18:57I'm not really sure why I did that.
- 18:59I just think it's good. It's good at cutting things,
- 19:01like, you know, I think when people are over long and you put
- 19:04a lot of text in it, it can all often strip away useless stuff.
- 19:10Sort of the summarization use case or yeah, yeah, I'm very
- 19:14excited about. Again, I think it's a great and
- 19:17we've sort of centered on this. It's great knowledge delivery
- 19:20tool like souped up sort of Google experience where Google
- 19:24was moving towards giving you the answer when you searched
- 19:27instead of directing you to a link like this like James was
- 19:30talking about with his child. Like I feel like it's great at
- 19:33sort of quick. Like I wish there was like a
- 19:37likely accurate score. Because I it does feel like what
- 19:41if one out of every ten of these is just like it's hallucinating
- 19:44and just making shit up. I'm I feel like humans including
- 19:48myself. I'm willing my.
- 19:50Daughter, My daughter's in trouble is, I know, but we're
- 19:52willing to accept like, oh, it's wrong 10% of the time.
- 19:55That's better than a lot of people.
- 19:57But then you just treat. You don't really know which ones
- 19:59were wrong, you know? So until we get burned.
- 20:02Yeah, the confidence score would be helpful.
- 20:03Yeah, I definitely have not never asked GPT for help with my
- 20:06daughter. Cause to your point, like if I'm
- 20:09like playing Russian roulette where it's like, oh, like like
- 20:12why that attendity? This is just wrong and I'm going
- 20:14to like take that away for my childcare.
- 20:17It's like, not great. Not a good, not a good idea, I
- 20:20mean. I like to think that I have some
- 20:22higher level filter of if it tells me to do something insane.
- 20:26I hope so, yeah. I mean, it's pretty good at
- 20:30plausible bullshit, though. I mean, that's the thing like
- 20:32it's it's really good at making wrong things sound plausible.
- 20:36All right, let's get out of our narcissism and think about how
- 20:41other people are using it. What?
- 20:44I mean, Max, I think you've made it clear, like students are the
- 20:47answer. Students are using this the
- 20:49most. And we you can talk about that a
- 20:50little bit more. And I totally agree.
- 20:53I think we all agree on that. But yeah, what professions do
- 20:57you think are using this the most?
- 20:59What types of people? And then we'll go from there and
- 21:02sort of what we think, who will be using it soon.
- 21:06It's pretty clear if you look at educational research that
- 21:09private tutoring is like the best possible thing you can do
- 21:13for learning in almost any context.
- 21:15Like one-on-one tutoring in like every research study that has
- 21:18ever been put to the test just absolutely destroys.
- 21:21Like the classroom environment or self-directed reading or you
- 21:26know, yadda yadda yadda what any, any sort of way you can
- 21:28think of to sort of teach people like one-on-one Private tutoring
- 21:32just kills for obvious reasons. I mean, right?
- 21:34You get personalized, you know, follow up feedback, custom
- 21:37lessons, all that good stuff. So you have to imagine, you
- 21:41know, if you could automate to some degree the learning process
- 21:44using, you know, a generative AI tool, whether or not that
- 21:50replaces teachers or that's just something teachers can use as
- 21:52leverage to really provide personalized feedback and lesson
- 21:55plans and and instruction to children.
- 21:58I just think that to come back to my initial analogy around
- 22:02this being like the industrial revolution for the mind, I
- 22:04really think we could see like insane gains in the quality of
- 22:08education over the next decade or so.
- 22:10Which is not, I think, with the intuition that a lot of people
- 22:14have, it's like, oh, it's helping them cheat.
- 22:15They're unplugging their brains. I think it's good if they're if
- 22:19the full experience is like interacting with the AI and
- 22:22going back and forth. And especially, you know, if the
- 22:26AI is set up to actually instruct so it knows when.
- 22:30To be withholding and not just like give every answer, that
- 22:33feels very powerful. But I do think the threat of
- 22:37it's easy to pretend to learn is like a real one and I'm I'm open
- 22:41to it going either way. I don't know.
- 22:43I mean this is sort of the Wall-e, you know like are people
- 22:47just going to be like laying back as the machine sort of
- 22:50taking over, but I think that might be slightly over indexing
- 22:53on like higher education, right? I think that if you think about
- 22:57what you learned up through almost like 8th grade right, a
- 23:02lot of it was like math, like addition, subtraction,
- 23:05multiplication and basic algebra, right?
- 23:08Phonics. Like reading, right?
- 23:10You know, I think reading comprehension can also be tested
- 23:13in person, You know, science. Just basics of the sort of core
- 23:19curriculum of elementary and middle school.
- 23:21I don't think like necessarily is going to be substituted for
- 23:27by like, oh, I need to dump an essay prompted to ChatGPT,
- 23:30right. I mean I think like, I think
- 23:31ChatGPT is obviously super good at the like give me a 5
- 23:35paragraph essay about Jane Eyre and like what the various
- 23:38symbolisms, you know in chapter four are or whatever, right.
- 23:41It's incredibly good at that. But that's sort of like a pretty
- 23:45narrow subset of what I would consider like almost all
- 23:48education all the way up to the college level.
- 23:50So I think that if you can give people private instruction on
- 23:54how to add you know or like you know really basic phonics and
- 23:58reading comprehension and grammar and sentence structure
- 24:00and you know science, geology, whatever, like all this stuff
- 24:03that you learn especially under the 5th or 8th grade level, I
- 24:06think it's hard to imagine we couldn't make really, really big
- 24:09gains in that area. And you know especially if you
- 24:11look at the United States where we fall short, a lot of it's
- 24:13just like making sure everyone reads really well by a certain
- 24:16age, right. And and I think AI could be
- 24:18pretty helpful there, particularly with voice and
- 24:20audio as like a multi modal experience, not just text.
- 24:24Don't offer a provocative one, I think.
- 24:27Have you read the book Bullshit Jobs or like this class Class of
- 24:31Tasks? Read a review of it, yeah.
- 24:33Like, I just missed it out of hand.
- 24:35No, I mean I would think you'd be sort of you'd buy it.
- 24:39I mean the idea is, you know that there are a ton of white
- 24:42collar jobs mostly that exist to sort of just move things around,
- 24:47don't create real value like a lot of, I mean, corporate,
- 24:51corporate law. I think is like a key one some
- 24:54of HR you know just a lot of where where is the real value
- 24:59versus just imposing a tax on a system that both sides have to
- 25:04pay. I mean to I think sort of not
- 25:09top tier marketing necessarily, but I think this sort of low
- 25:13tier marketing is obviously a case that I put in this bullshit
- 25:18jobs category where a sort of. And it's somewhat derivative.
- 25:21You know, like you really don't need to.
- 25:24You know, hopefully artists are coming up with many of the
- 25:27original ideas and so you're trying to sort of be in touch
- 25:30with culture and pull something. What do you mean exactly by low
- 25:34tier marketing? I guess 'cause.
- 25:36Like I just mean like you know people or you know, just I think
- 25:39obviously like, I mean I think there will be an AI Super Bowl
- 25:44ad, but I put that in sort of a. Sure, it will be like the
- 25:47creativity of it, but you know, that's going to be reviewed by a
- 25:51human, right? I mean, like if there's an AI
- 25:52Super Bowl. But I'm just, I'm saying like,
- 25:54OK, somebody's like I'm putting up a website for our company and
- 25:57we need to explain all this stuff and you read all that copy
- 26:00and it's just like this is like, no, nobody paid somebody who's
- 26:03like great at writing to do something novel.
- 26:05They're like, what does website copy look like?
- 26:09What does our company do? Let let's create website copy
- 26:13for. Plus and so then, if that's your
- 26:16standard of what you want, you just want to have a website that
- 26:18has things that are not wrong. I think ChatGPT and other
- 26:23similar products will be able to deliver that.
- 26:25Yeah, I think you're right. I just would push back.
- 26:28So I probably would have like fully agreed with you like five
- 26:30years ago, but now, now, now having now having run a start up
- 26:35and and experience a lot of these things that people
- 26:37consider to be bullshit chops. I think the thing that you maybe
- 26:41don't realize having been like close to the metal on like
- 26:43putting up a website or whatever is there is actually quite a lot
- 26:48of people in the world who will sell you the service of putting
- 26:50up a generic website that will do a pretty bad job.
- 26:54Like a worse job than you're describing as a sort of like you
- 26:58know an AI quality or a low tier quality output, right.
- 27:02People will like, literally not even deliver at all.
- 27:06Or they will deliver something that's like grammatically just
- 27:09incomprehensible. Or they will deliver something
- 27:11that's grammatically comprehensible but really quite
- 27:13bad and does not describe what the product does.
- 27:15So you will encounter like just a staggering number of services
- 27:19people that this. Could cover is much larger than
- 27:22right. I'm saying that, but I'm saying
- 27:23in the end like the leverage point ends up being like the
- 27:26person who can look at the really bad version of it and be
- 27:30like well actually there's like a mid tier version of this that
- 27:33we could just get an AI to do right and and go get the AI to
- 27:37do it and then decide side that that's actually good enough for
- 27:39for showbiz to go on the website, right.
- 27:41So I'm just saying like I I think that I think that that the
- 27:46humans can always sort of move up the stack I guess you know
- 27:49until we get super intelligence or whatever.
- 27:51But like where the the key the key job is deciding what is like
- 27:55good enough output, not like creating good enough output in
- 27:59many cases I guess I would say. I mean like and yeah, a core
- 28:03question to me is like from what you're just saying there.
- 28:08Is does AI make dumb people smarter or smart people higher
- 28:14leverage? Like in the most brutal way?
- 28:17Like, is it? And it's not just smart and
- 28:19dumb, but like in your case, it's like, does it help the
- 28:21person with taste do a ton of jobs, or does it help have the
- 28:26person with sort of poor taste or like bad execution skills?
- 28:30But maybe they know it's not good and therefore they can do
- 28:33better, like I think a lot of the takes have been.
- 28:36It helps sort of the the person who needs more help, yeah.
- 28:40But there's also this argument that it's scaling people.
- 28:42I think it's different on taste, I guess, But I do think the
- 28:46early academic research that's coming out is showing that.
- 28:49I think they gave it to lawyers who were in, in law school,
- 28:52basically, and they let them use it on their tests essentially.
- 28:56And it did. So they had a curve, you know,
- 28:59everybody kind of knew what their grades were going into the
- 29:02AI assisted test basically. And it did pull up the bottom
- 29:05half of the curve quite a lot and then the top half of the
- 29:08curve maybe a little bit, but not a huge impact.
- 29:11So at least in the law and I don't know if this will be
- 29:14different and more, you know, different types of fields, more
- 29:16creative fields versus more structured fields.
- 29:18It's it's a little bit hard to predict.
- 29:19But in the case with the law where there's sort of clear
- 29:22right and wrong answers and there's a value in having
- 29:26massive text digestion machine essentially which is what these
- 29:30LLMS are, right. It seemed like it was helping
- 29:33the bottom of the curve substantially more than it was,
- 29:35giving like a ton of leverage to the top of the curve.
- 29:38I would also add there. This kind of reminds me of an
- 29:41analogy with GPSI. Forget who's who said this
- 29:45originally, but we went from an. An era where we were to
- 29:51navigate, we had someone maybe in the passenger seat just
- 29:54pulling out maps and you know, they could actually help
- 29:57navigate in natural language and you know get you to where you
- 30:00wanted to go or you had to pull over and you know pull out the
- 30:04maps yourself. And then we went to GPS and the
- 30:08navigation at first got I was actually worse than having
- 30:11another person just look like a human look at the maps.
- 30:14But it did make. On the whole, like a lot of more
- 30:18convenient trips, you could just, you know, throw on the
- 30:21navigation. Similar with early call centers.
- 30:25Like it was nice to when you originally it was all humans and
- 30:27you just talk to them and you get customer support.
- 30:30And then with call center automation, many, many more
- 30:33firms could create call centers, but they were way worse, Like
- 30:36you couldn't get the answer that you wanted and that, you know,
- 30:41was frustrating for consumers. And I think we're going to see
- 30:43something very similar here with AI that.
- 30:47Just like Max was saying, like AI can write spam emails.
- 30:49Like for sales. Like those emails are going to
- 30:52be way worse than a human writing a spam sales e-mail and
- 30:55you're going to get way more of them.
- 30:57So overall, our lives as startup founders for now might get worse
- 31:01because we're getting 10 times as much spam e-mail and trying
- 31:05to sell us stuff, but on net. Maybe that's good for the
- 31:09economy or something like more firms can use spam e-mail
- 31:13tactics to sell products, so. I I think that a lot of times
- 31:17that the the technology creates the ability to scale things that
- 31:21weren't scalable before but on net makes that experience with
- 31:25the customer worse. And I think we're going to start
- 31:28seeing that and that that might because it's so much cheaper to
- 31:32do so much. That might go for website copy
- 31:34too. Like it'll be way easier to
- 31:36throw up a website on the on the Internet, but the average
- 31:39website quality will go down because.
- 31:42The writing will be worse and anyone will have the ability to
- 31:45put a website up and code it with AI.
- 31:47And yeah, I I think I don't expect that to like raise the
- 31:51quality of the average website. That ties into Max's tutor
- 31:55point, right? I mean it.
- 31:56It's democratization to make something that was expensive,
- 31:59private tutoring cheaper, and possibly the original quality is
- 32:04lower. But way more people can access
- 32:07it. One one last point on the
- 32:08website topic, actually there's this a very funny story I
- 32:11experienced where I talked to investor and he said one company
- 32:14in his portfolio had used GPT to create 8 million websites, 8
- 32:21million web pages essentially that were all meant to
- 32:24essentially attract Google search traffic essentially.
- 32:28And then they would use the how many clicks those various pages
- 32:33got to like decide what the next thing is that they should work
- 32:36on or the next thing they should create as their company.
- 32:39So essentially like, yeah, they were very early on this.
- 32:42Like we're going to just like absolutely, just like carpet
- 32:47bomb any keyword imaginable that is remotely related to the
- 32:52business and then figure out where the user traffic is and
- 32:56essentially use it as like honey pot or whatever to see where the
- 32:59traffic is. And then that's what we're going
- 33:01to build next. And so I think your idea
- 33:03specifically of just the number of spam websites going through
- 33:06the roof clearly is is either happening, going to happening.
- 33:10Yeah. I mean is is going to be a huge
- 33:11part of and and Google search is really going to have to deal
- 33:14with this. Like, how do we deal with the
- 33:16fact that it's now become, you know, three orders of magnitude
- 33:20cheaper to create a crappy website?
- 33:22And how do we filter through that with Google search?
- 33:25I feel like a theme that's emerging on this podcast already
- 33:27is the idea that AI is going to be at war with itself, Like it's
- 33:32going to bring improvements, but it's also, it'll make it easier
- 33:35for spammers. But we have to hope that Gmail
- 33:38also gets better at filtering them out and that to some degree
- 33:42there's going to be sort of a back and forth there and that if
- 33:45if that works out well. But you know that will, that
- 33:48will be for the best. I mean even spammers.
- 33:51I mean in James's GPS analogy. Like, the end point is that they
- 33:55have to write good spam that is compelling to you, and that that
- 33:58that's a winner rather than just living in sort of the shitty
- 34:01spam era forever. Yeah, but I think the spam
- 34:05e-mail example is is important here because.
- 34:09They can annoy 99% of people who won't convert, but if it helps
- 34:13convert any extra 1% of people, then that's a win for that
- 34:17company, right? It could be 99.99.
- 34:20I mean it could be one out of it could be one out of 10,000 of
- 34:22these emails convert or 100. Like on spam.
- 34:25Eric, you're naively optimistic. Sweet.
- 34:29All right, we. I am interviewing Joshua Browder
- 34:33at Do Not Pay. So stick around and give that a
- 34:37listen. Welcome, welcome.
- 34:42Joshua Browder, CEO of Do Not Pay, welcome to the Cerebral
- 34:46Valley Podcast. Thank you Eric for having me.
- 34:50Yeah, I mean, I I, you know, you've had such a great journey.
- 34:54I think some people might be familiar, you know, just from
- 34:57Twitter following it. But can you talk about, you
- 35:00know, I want to end this conversation, really focus on
- 35:02what do not Pay is doing in AI, but like.
- 35:05You've got such a great journey. If you could just give a little
- 35:08bit of the sketch of how you get into this world of helping
- 35:11people fight, you know, bills, traffic tickets, everything,
- 35:16yeah. And then how, how that sort of
- 35:18evolves into this sort of generative AI world?
- 35:21Sure. So I'm the founder of the
- 35:23company called Do Not Pay and Do not pay as an AI legal agent.
- 35:27So. There are so many areas in life
- 35:29where people are being ripped off, from parking tickets to not
- 35:33being able to cancel their subscriptions, to junk fees, and
- 35:38no one has time to wait on hold for five hours to argue over
- 35:41$12.00. And so that's a really good job
- 35:43for AI and software. I started the company six years
- 35:47ago with templates. When I moved from England, I got
- 35:50a bunch of parking tickets. I was a terrible driver.
- 35:54And I realized if you know the right things to say, you can get
- 35:56out of your tickets. You're you're speeding or you're
- 35:58just putting, you're putting your car wherever or.
- 36:02In in the UK test they don't test how to do parallel parking,
- 36:06so I wasn't particularly good at that skill, especially on the
- 36:09other side of the road. Hey, listen, I grew up in Macon,
- 36:11GA and you did not. I failed the parallel parking
- 36:15part of my test. I didn't know you only had like
- 36:16three tries to try and get it in.
- 36:19But I didn't. You know you could pass without
- 36:21without succeeding on that portion anyway.
- 36:23Yeah, so. I think in the UK test they do
- 36:27have it, but it's random and I got another random one that was
- 36:30really easy. I think my one was a three-point
- 36:32turn, which I can do in any case.
- 36:35So I created the first version of Do not Pay.
- 36:37I looked at the top 12 reasons from Freedom of Information Act
- 36:40requests. A wide parking tickets are
- 36:42dismissed. Built templates around that and
- 36:45really just built it for fun. And I could never have imagined
- 36:48that. America and the UK and the world
- 36:50would hate parking tickets and it would go viral.
- 36:53And this made me realize that this idea of automating consumer
- 36:56rights is bigger than just tickets and I should work on
- 36:59other areas of fighting back. So.
- 37:02Fastball parking tickets thing. 20/15/2016 so I was a freshman
- 37:07at Stanford. And at first, it's not automated
- 37:09at all. Like you're doing it personally
- 37:11yourself and people are like asking you for favors.
- 37:13Is that what gets you into sort of automating things in the 1st
- 37:16place or had you been a sort of? In other domains trying to to do
- 37:21automated stuff. Yeah, just from a time
- 37:24perspective. People were asking me and I was
- 37:26in Google Docs back then copying and pasting letters.
- 37:29And I just built it, made it automated because too many
- 37:32people were asking me and what a rumour had spread that I was the
- 37:36guy who could help people with parking tickets.
- 37:38And that's not a good rumour to have about you because everyone
- 37:41will bother you. And so I just built it out of
- 37:43necessity, really. When does it become a company?
- 37:48So in 20/16/2017 I was very lucky.
- 37:52Andreessen Horowitz invested 1,000,000 into the precede
- 37:56rounds. I still didn't know have any
- 37:59business model or anything like that, but I guess it became a
- 38:02company legally at that point. But it was really on 2019-2020
- 38:06that I actually started charging.
- 38:08So I just saw it as a free public service until I started
- 38:11charging. I did it not to do a start up or
- 38:14make money, but because I like helping people fight back.
- 38:17I like think it's so unjust all of these big corporations and
- 38:21governments and so I was more mission like activism driven and
- 38:25that's proven by like 5 years of no business model.
- 38:29A good crusade is important. You know what?
- 38:31You can get investors to give you money to go on a crusade.
- 38:34What's better than that? The I mean the the rise of
- 38:40ChatGPT Foundation models. What I I I I can guess that a
- 38:45key value for you is that it can generate like demand letters on
- 38:50its own. Like what have been the pieces
- 38:53of large language models that have been most useful so far for
- 38:57Do Not Pay? Really, it's synchronous
- 39:00responses. So in the realm of what we can
- 39:03fight back, parking tickets is asynchronous.
- 39:06You generate A latter, send it off, and wait three weeks.
- 39:09But there are some things that you need to do synchronously.
- 39:11So for example. Bill disputes.
- 39:14So one way you can dispute your Comcast bill is you log in and
- 39:17you go into online chat and you negotiate with them for an hour
- 39:20on online chat. That is a great job for AI
- 39:23because it doesn't give up. And better yet if the agent
- 39:26denies the request you can just start a new chat and that's what
- 39:29we have bots doing. So they just keep trying, keep
- 39:32pushing until we get a bill reduced.
- 39:35So synchronous is 1 and not giving up.
- 39:37I think are are two and also being dynamic.
- 39:42But not the letters itself. Or are are you using it for sort
- 39:45of demand letters? So the contents of the latter we
- 39:50want to kind of free craft, but really it's about the
- 39:54communications out surrounding that.
- 39:56It doesn't really differ from person to person.
- 39:58So for example, the top reason why people get out of tickets is
- 40:02the poor signage. We've all seen those memes on
- 40:04Twitter of Like 5 signs saying the opposite thing.
- 40:07On top of each other or a tree covering the sign.
- 40:10So the substance of the defence doesn't really change with AI,
- 40:13but the way you can communicate it does.
- 40:16You have a person who has to figure out the the argument,
- 40:21yeah, to build the kind of system and then and then AI can
- 40:26go back and forth communicate, yeah.
- 40:29And I I'm not, you know, I guess for context for the listener,
- 40:32you had this whole you know, we'll be your lawyer period or
- 40:35can you talk about? Where sort of the legal and
- 40:38policy realm has been something you've run into and sort of the
- 40:42quest to operate as like an attorney, basically and the
- 40:48protectionism that you found in that crusade.
- 40:53Yeah, so consumer rights is completely underserved.
- 40:56There's not a lawyer who will get out of bed to help you with
- 40:59your parking ticket, which is our first use case.
- 41:02Or Dell disputes or any of this, especially in flight Wi-Fi
- 41:05refunds which are literally $30.00 so.
- 41:09When Oh yeah, I've had that where the Wi-Fi goes down and
- 41:12then you have to like you're like you didn't give me a
- 41:13service, like how can you charge me a bazillion dollars And then
- 41:17it's hard to ask. For the refund, because you
- 41:19don't have Wi-Fi, it's ridiculous.
- 41:21Yeah, yeah, that that's a great job.
- 41:25There was a time for about two months at the beginning of this
- 41:28year where I wanted to actually bring ChatGPT into the
- 41:31courtroom, and I got a lot of push back from that from from
- 41:35the judges and regulators. Because which I love.
- 41:37To be clear, our supporter 100% on this podcast.
- 41:40Anyway, go ahead because it's nice to help people online, but
- 41:45if you can actually go into court, you can go into even more
- 41:48advanced areas. And they push back so hard that
- 41:51I thought, I'm not getting any push back with consumer rights.
- 41:54Everyone hates Comcast. Even the lawyers do, and so it's
- 41:57best to just stick to the undeserved area to help people.
- 42:01The provocative fight, you know, like I'm going to be your
- 42:03lawyer, you know, you get PR attention out of it, but then
- 42:06yeah, you bring down the regulatory heat.
- 42:09So there's a double edged sword there.
- 42:12I mean I get the the philosophical issue here or like
- 42:15they're they're just like the sense that there are systems at
- 42:20war against the consumer, right. Like a Comcast, you know, can
- 42:25think about the process and then you were an individual, not
- 42:28teamed up with other consumers. Sort of fighting with your arms
- 42:31tied behind your back and and similarly, you know, I think
- 42:35part of the threat to the legal system is that this would expose
- 42:38that it's sort of a system of repeat actions that happen over
- 42:42and over again and slight adjustments and even lazy ones,
- 42:46you know, lazy and that you know that it's a computer, not like
- 42:49some person could have a big, big change.
- 42:52So talk about you know how this is sort of going up against
- 42:56these systems on behalf of the consumer.
- 42:59My most optimistic view of AI is it will end up patterns.
- 43:03So right now we have this problem in Society of
- 43:06concentrated benefit but spread out harm.
- 43:08So what I mean by that is Planet Fitness can make it super hard
- 43:12to cancel and charge 10 million people $30.
- 43:16They make a 300 million but the average person is only getting
- 43:21$30 taken out of their account. So they find it very difficult
- 43:24to fight back. And with Planet Fitness
- 43:26specifically in most gyms, you have to actually sign a physical
- 43:29legal latter and mail it to cancel.
- 43:32And so all of these hoops that that these big companies made
- 43:36you jump through AI can jump through it because it doesn't
- 43:39require a salary and doesn't have anything to do.
- 43:42So you can get it to do that. And then all of these kind of
- 43:46systems that prevent people from take fighting for their rights
- 43:50can can go through. Another example is privacy.
- 43:54There's an amazing law in California, the California
- 43:57Consumer Privacy Act, but it's largely been a failure because
- 44:00no one really has exercised their rights.
- 44:02The the law says that you can request to delete your data or
- 44:05not sell your data, but no one has time to fill in all the
- 44:08forms. We have AI going send 1000
- 44:11requests to every data broker on behalf of consumers.
- 44:15So I think AI will really help level the playing field and give
- 44:19power back to ordinary people to fight this concentrated benefit
- 44:22problem. So you can you talk about like
- 44:24the mechanics of using AI, like is this ChatGPT or what are you
- 44:29finding right now that's the most useful to you?
- 44:33So we're mainly using the GPT for API.
- 44:37So for example, with Comcast, what's interesting is they're
- 44:39also using AI, so the two AIS are sometimes chatting with each
- 44:43other to negotiate. They're already a war, yeah,
- 44:45yeah. And it feels like we're making
- 44:48so much progress every few weeks and months with the new
- 44:51releases. So the two.
- 44:53So the biggest was when GPT 3 upgraded to GPT 4, it became a
- 44:57much better negotiator. So with GPT 3, Comcast would be
- 45:01like, OK, I'll give you $20 off your bill.
- 45:05And GPT 3 would say, yeah, that sounds great.
- 45:08Thank you so much. Now GPT 4 says, no, that's not
- 45:11enough. I want $100 or I'm going to
- 45:13cancel right now. And it does these high stakes
- 45:16negotiations that I don't think previous AI models were capable
- 45:20of. And then the thing I'm really
- 45:22excited about now is the multi modal stuff.
- 45:25AI is not useful unless it can interact with the world and
- 45:29being able to send it images and all sorts of different media
- 45:33PDFs as well is really helpful, especially in the legal context.
- 45:37So you can imagine it looking at like a parking Bay and saying
- 45:40that parking Bay is not to code, which is why I should get out of
- 45:43my ticket. So that's all the stuff we're
- 45:45working on. Right.
- 45:46Now you sort of explained this in the answer, but multimodal
- 45:48just like putting in text and images into the same system, not
- 45:52totally separate system so that it could like you're saying,
- 45:55process a photo and then respond in text to it.
- 45:59How much are you saying Open AI is the expert at ChatGPT like
- 46:04they're going to keep improving it.
- 46:05A general language model is just going to be better than anything
- 46:09we're doing versus the idea that like you are now specialized in
- 46:13this. Use case of like consumer
- 46:15defence and therefore you should like train a model on those
- 46:20interactions. Like how do you think about how
- 46:22much to sort of fine tune or build your own language model.
- 46:27I I think the the kind of sweet spot is you have these kind of
- 46:32big commercial models and then you fine tune them.
- 46:34So what we do is we feed it like successful cases and and
- 46:38existing letters that we think are good.
- 46:41Any answers, like every time they're like, here are the
- 46:43models of the good ones. And then, yeah, yeah, like here,
- 46:47here's an example. So when we were doing our
- 46:50courtroom stuff, we didn't open. AI is so valuable to our
- 46:54business, we didn't want to get banned by doing this
- 46:56controversial. Stunt interesting.
- 46:58And so and so we said we were going to use open source model.
- 47:02So we had to go deep on the open source ones to get it to work.
- 47:06And it feels like it's like 75 percent, 80% of the way there at
- 47:10this point, the open source models.
- 47:12So I'm very confident that they'll continue to improve and
- 47:16it will always be at that level. So I'm a big believer in open
- 47:20source and and there are going to be things if you want to have
- 47:23a big impact, there are going to be things that are quote UN
- 47:25quote against the rules. So you you will have to use an
- 47:28open source model to do some interesting stuff.
- 47:31You're fighting for the consumer, but you're there's
- 47:33already sort of the status control, I mean under it makes
- 47:36sense like you know. Open AI doesn't want to be
- 47:39exposed to your sort of niche like principled fight, even
- 47:43though they probably should. That's super interesting.
- 47:47How? Which?
- 47:48Which open source models have you been using?
- 47:50Like which ones do you see the most promising matters?
- 47:55One is good we in January, we're using GPTJ.
- 47:58Yeah, GPTJ is is good at holding a conversation if you kind of
- 48:04fine tune it enough so that that's what we were using to
- 48:07power the courtroom stunt. And is it expensive on Chachi BT
- 48:13just like to build a business around it, Like how how much are
- 48:16you paying? We're paying 10s of thousands
- 48:20every month. I would say that for us it makes
- 48:23sense. We're a subscription model.
- 48:24We fortunate to have a lot of subscribers and a subscription
- 48:27is a great business model. There are some things everyone
- 48:32that they don't do these great demos on Twitter, but we're
- 48:34thinking as a do not pay team, this would never work if it
- 48:37wasn't in the do not pay business model.
- 48:39So one example is every few months someone comes up with an
- 48:42idea of ChatGPT, scanning websites for terms and
- 48:46conditions, whether it's good terms and conditions or bad.
- 48:50And probably each scan maybe costs two cents or something
- 48:55along those lines, one or two cents.
- 48:57I don't think that's something consumers would pay for.
- 49:00Imagine if you're being charged $0.02 every website or you visit
- 49:03so that. Would be a great browser.
- 49:05Feature like Chrome's like, you know, it's like how with a
- 49:07credit card you get a bunch of features you could imagine
- 49:10someone paying for like a browser.
- 49:12If it's like, one of the things we do is we say this has really
- 49:15aberrant like terms and conditions.
- 49:17So we should, you know, that would be an interesting company.
- 49:20Yeah, it it would. But so we have 200 products and
- 49:24from my experience, it seems like consumers can most about
- 49:28getting money back. That's the number one thing
- 49:30privacy they do, they do care about.
- 49:32There are certain niches of people, but the mass market
- 49:35people are just trying to get by every day and so it it would
- 49:39have to be very cheap for them. Do you get a?
- 49:43Have you found any ways to get a percentage of the money you save
- 49:45people? We think it's unfair.
- 49:49So Equinox is $300.00 a month. We help people cancel Equinox
- 49:53subscriptions. We could.
- 49:55There are some companies that say we've helped you cancel
- 49:58Equinox subscription. We've saved you $3600.
- 50:01We're going to take 20% of that. And we believe that business
- 50:04model is not actually consumer friendly.
- 50:07We'd rather and we have 100% success rate with cancelling
- 50:10Equinox. We'd rather just charge the $13
- 50:13a month subscription. We'd rather just charge the
- 50:15subscription. How would you judge like the
- 50:18intelligence of ChatGPT and the open source models based on what
- 50:22you've seen you you've talked about?
- 50:25You know, oh like, wow. It can really start to like,
- 50:28make threats and like, say, OK, this is a final, like, you know,
- 50:31you see a ton of this now. Like, does it feel like Super
- 50:34Rodent has the same playbook, or does this feel like something
- 50:38with intelligence to you? It's extremely intelligent.
- 50:43From the legal standpoint, GPT 3 was not so impressive.
- 50:46It would make very basic mistakes, like mix up the
- 50:48defendant and the plaintiff and all of this stuff.
- 50:52GPT 4, those mistakes have disappeared.
- 50:55My biggest worry with it is just like just like humans, it's
- 50:58almost too intelligent, so it gets on the verge of
- 51:01manipulative, so those who can use it face liability from it.
- 51:05Potentially lying. I think there's a difference
- 51:08between hallucinating and lying. So with lying, if the goal in
- 51:12the prompt is like I want to get a discount on my bill, it will
- 51:16say things like I've had four outages in the past week.
- 51:19Oh no, that's not true. Oh my God.
- 51:23So we've had to kind of build guard rails about around it in
- 51:26the prompt. And also we even have another
- 51:28kind of pseudo AI model on top of it and the prompt says like
- 51:31stick to the facts, stick to the provided information.
- 51:34So because we have liability, we we can't have it lying on behalf
- 51:37of people. That's not good.
- 51:39We're responsible for that, so. That's fascinating, yeah.
- 51:43The distinction between hallucinations where it's just
- 51:47it's just confused basically in lying where it knows that it's a
- 51:50tactical. Move.
- 51:54Yeah. I mean, I mean there is, you
- 51:56know, there's an argument that it doesn't have enough
- 51:58information. Like you know it because it.
- 52:01Right. You know, that you could face
- 52:02like legal consequences. Yeah, that's a fascinating one.
- 52:09Are there cases, I mean, people, consumers, do lie to these
- 52:12companies sometimes when they try and get out of it?
- 52:14Are are there cases where you're OK with where that's allowed or
- 52:18you basically have to be 100% factual?
- 52:23I I think you can be aggressive so you can say I'm not happy
- 52:26with the service, it barely works, things like that.
- 52:30But you don't want to be opinions.
- 52:33I'm going to cancel tomorrow and maybe they're the only Internet
- 52:36company in the area. So you can't cancel but so, so
- 52:40those things I'm comfortable with.
- 52:42But right specific concrete details we we don't feel
- 52:45comfortable doing it. We we have our own we have some
- 52:48of the best lawyers that do not pay that we were hired
- 52:51ironically and and we're we're trying to stay in compliance.
- 52:54So it will help us a lot. Are you going to go back into
- 52:57the legal realm at any point or are you you sticking to fighting
- 53:00companies or I guess you? Fight companies in the legal
- 53:05realm, Yeah, in the legal. There's a lot of interesting
- 53:08stuff that we're working on now that's definitely in the legal
- 53:10realm, but it's focused on helping consumers with their
- 53:14everyday issues. So one thing that I'm really
- 53:16excited about is everyone has money lying out there from class
- 53:21action settlements. There's like $20 here. 50
- 53:24dollars there, because if you're a customer of Macy's a few years
- 53:27ago, so we've built like a bot that will go through your
- 53:30emails, figure out which companies you're owed class
- 53:33action settlements for, and just claim the money.
- 53:36So that's squarely in the legal realm.
- 53:38But once again, it's underserved and it's just helping ordinary
- 53:40people. We're not trying to help defend
- 53:42people from murder, although that would be.
- 53:45We're not Sam Bang Bun Freed's trials today.
- 53:47But we can't help Sam Bang Bun Freed.
- 53:49But we can help ordinary people who are don't have these
- 53:52complicated issues in their lives.
- 53:55How much? How much do you use AI, Chat,
- 53:59UBT any of this in your personal life?
- 54:02Like you're someone who likes to automate your problems.
- 54:05To what extent have you found it useful outside of your company?
- 54:10My company is just me trying to scale myself.
- 54:14And so almost all our products start with either me or another
- 54:17team actually doing this in their personal life.
- 54:21So one example is I built a bot that phoned up Wells Fargo on my
- 54:26behalf, and I actually created a voice model of myself and it
- 54:30negotiated with Wells Fargo to get some bank fees back.
- 54:33We decided not to make that into a product for a variety of
- 54:35reasons, mainly to actually make it a working product.
- 54:39You need to record what they're saying and transcribe it so the
- 54:42AI knows what to say back. And there's lots of federal
- 54:45wiretapping laws and state wiretapping laws and so we
- 54:50couldn't make that product, but it was helpful for me.
- 54:52Another example is I using Plaid, I connected chat TBT to
- 54:56my bank account and I got it to come up with different ways I
- 55:00could save money so that people do not pay products around that.
- 55:03So I'm definitely a Guinea pig everyday for these AI
- 55:06technologies, which is I love. I like to say I live the do not
- 55:09pay lifestyle. I'm always fighting back.
- 55:12Is it always, it feels like ChatGPT looms super large.
- 55:16I mean there is this question of whether open source models or
- 55:20Google Gemini or what, you know, whatever will be sort of
- 55:24competitive. Are are you sort of saying right
- 55:26now that it's chat GPTS to lose? I would say that the the only
- 55:33reason they would lose is if they restrain themselves too
- 55:35much. I think they're under a lot of
- 55:37pressure. I know lawyers are evil and very
- 55:39greedy and I think that a lot of lawyers are actually going off
- 55:41the ChatGPT right now and they're saying you have to put
- 55:45this disclaimer and that disclaimer and stop doing this
- 55:47and stop doing that. And then the AI will just not be
- 55:49useful and then they'll be like some foreign company where the
- 55:53US users can use the API that might overtake it.
- 55:57I I think we're on the verge of like society collapsing from all
- 56:01the new regulations and lawyers, and so I do think there's a real
- 56:05danger around that. Yeah, Just there's sort of
- 56:08regulatory legal state gets to it that creates sort of an
- 56:12opening for the unregulated open source world.
- 56:17So I I can give you a concrete example actually.
- 56:20So one when I used ChatGPT was I actually connected my credit
- 56:25report to ChatGPT and I asked it to like dispute things on my
- 56:30behalf. And then I recently tried to do
- 56:33it again, and then I got all these disclaimers and it
- 56:35wouldn't even let me. So they're they're definitely
- 56:38trying to all these high, high risk areas 1 by 1.
- 56:40They're kind of foreclosing on them and stopping useful
- 56:45implementations. Do you, you know, do you think
- 56:48it's like white collar workers are better at being Luddites and
- 56:53protecting their jobs? That, and they're happy to let.
- 56:56Sort of factory workers jobs be automated.
- 56:58It's like lawyers protecting their jobs or it's just sort of
- 57:01paranoia or what do you think motivates this?
- 57:06The white collar workers are definitely going to put up a
- 57:08fight with the Actors Guild. I know in their, I'm sorry, the
- 57:13Screenwriters Guild, they put in their contract that they can use
- 57:16the AI, but the studios can't, which seems seems a bit
- 57:19unbalanced. Lawyers are the ones who write
- 57:22the rules and so of course they have an advantage.
- 57:25I know a lot of state legislatures are passing AI
- 57:29laws. I think Utah was was the latest.
- 57:31They have a few AI bills coming up.
- 57:33It feels to me like we are going to get to a place where we each
- 57:36have our own sort of bots. Like the advantage being it
- 57:40knows the facts, like I trusted to be in my bank account.
- 57:44It sort of has the the data advantage on me that it has the
- 57:47full picture, sort of train the personality around what I how I
- 57:51like to interact with people, you know, a bazillion reasons.
- 57:55Do you see that as a threat to you?
- 57:58You know you're in some ways like specialized around a set of
- 58:01tasks. But if if you're going to be,
- 58:03would you think you'll be competing against a different
- 58:06model where it's all about the person and therefore they do a
- 58:09lot of stuff. We've we've thought about this
- 58:12for a while because consumer rights what what happens if
- 58:17Google just build an assistant that do all your consumer rights
- 58:19for you and and you're right it should be just part of a super
- 58:22assistant. Fortunately we are so anti
- 58:25authority that we're willing to do things to help people that
- 58:28they love that big companies won't.
- 58:31I imagine that Comcast has a huge advertiser on Google and so
- 58:35they're not trying to be too adversarial and so we think of
- 58:38things like adversarial AI. So plug insurance is a great
- 58:42example. When ChatGPT released plug
- 58:44Insurance, a lot of people when we thought is Comcast going to
- 58:48build a plug in to lower your bill or or Planet Fitness going
- 58:52to build a plug in to cancel your subscription from ChatGPT
- 58:55And I think the answer is obviously no because it's
- 58:58adversarial and so this adversarial kind of use case
- 59:01helps us. I I think what's going to happen
- 59:04is there'll be these niche application layer on top of
- 59:07these general assistant for these specialised areas that
- 59:10require either a different brand or some specialised expertise
- 59:15that Google is not willing to take on.
- 59:18Yeah, that that makes. A lot of sense.
- 59:20Do you have anything like sort of the general assistant or how
- 59:24much time are you spending getting a single AI with enough
- 59:28memory to keep track of all of your your information?
- 59:32Well, so we have chat bots that talk to our consumers and our
- 59:35consumers what expect jokes they they they talk to it about
- 59:39irrelevant issues. So we have thought a lot about
- 59:41the general use case and we want to make it entertaining.
- 59:45Eventually. We have a kind of we draw the
- 59:47line and it says I'm, I'm actually an AI legal assistant,
- 59:51so please can we like stick to that topic.
- 59:53But we we have a few jokes in there.
- 59:55But do you like when you're doing the Plaid integration with
- 59:59your bank account or credit report, is that all synced up?
- 1:00:02Like, do you sort of either through a prompt, say hey,
- 1:00:04here's everything we've done together in the past, or like
- 1:00:07try to buy extra memory in some way, use the API to like how
- 1:00:11much are you trying to, like, tie different tasks of yours
- 1:00:13together in one sort of prompt? Well, I guess there's two ways
- 1:00:18to do it. So you can either have a very
- 1:00:19large context window and I think anthropics right now is the
- 1:00:23biggest and so that's really exciting.
- 1:00:25Or you can build like these auto GPT implementations where it's 1
- 1:00:28task after another we we've gone the auto GPT route.
- 1:00:34So it's like incremental and I I think that's the solution, but
- 1:00:37hopefully there's like one day there'll be this AI with this
- 1:00:41massive contacts window and then that'll make things so much
- 1:00:44easier for us. With with the auto GPT, you know
- 1:00:46it can be. It feels like they need to have
- 1:00:49some response every time. It's like very hard to just say
- 1:00:53here's some information just say OK and then let me give you more
- 1:00:56information like is it a problem that it always seems to want to
- 1:01:00be completing some task with every answer if if it's very
- 1:01:06discreet and you've really kind of narrowed it.
- 1:01:08I I've seen it work. So we've used it.
- 1:01:11It goes on the government website and if if say you move
- 1:01:15address and you own a check and they can't the big companies
- 1:01:18can't reach you they just send it to the government and so it
- 1:01:20becomes an unclaimed money. So we've built an auto GPT bot
- 1:01:23that checks for unclaimed money and the reason we have to use
- 1:01:27auto GPT rather than just like selenium or define scraping is
- 1:01:31because the the web page changes and what we found is it can be
- 1:01:35useful and actually it's pretty successful at that use case.
- 1:01:39If you said to auto GPT order me a pizza the website will say
- 1:01:44like which topping do you want and then auto GPT will be stuck.
- 1:01:49So if a discrete task is good but the more open end, I think
- 1:01:52we are a long way from general AI assistant, maybe like a few
- 1:01:57years, which which is not that long, but it's not like
- 1:01:59tomorrow, right? Have you do you have you gotten
- 1:02:03a taste of where Chachi PD is headed?
- 1:02:06Or do you have you seen many beta products?
- 1:02:08Or what? Do you do you think the rate of
- 1:02:10improvement is going to slow down a lot after the current
- 1:02:14model? I I do.
- 1:02:17I think the current we we have a like a million applications
- 1:02:22waiting to be built. The Do Not Pay team is even
- 1:02:24swamped in our use case and so I think the next year or two is
- 1:02:28going to be really exciting with these AI applications being
- 1:02:31introduced to every product. Like, we're going to get better,
- 1:02:34but the models maybe aren't, well, not just us, but the
- 1:02:38whole, right, right. I think we'll build incredible
- 1:02:40things. You'll see sales tools, all of
- 1:02:42the stuff, customer service, phone lines, but it's just a
- 1:02:46statistical model at the end of the day.
- 1:02:49And what worries me is that I think GPT 4 is at the limit of
- 1:02:53the number of parameters. I'm not sure how many more
- 1:02:56parameters and data they can see then.
- 1:02:58They're sucked in everything there is to know, Yeah, they're
- 1:03:01sucked in everything that is. So I think there has to be a
- 1:03:04fundamental breakthrough with how these models are designed,
- 1:03:07which hasn't happened yet. I I think it will happen because
- 1:03:10there's so much money flowing into the space, as you know, but
- 1:03:14it's not like it's where it's secretly been done and just
- 1:03:18hasn't been released yet. Maybe that's above my pay grade,
- 1:03:21but I I don't know of that and I don't think that's happened.
- 1:03:24What what do you make of the politeness of Chachi BT or I?
- 1:03:28I find one of the big weaknesses, just like how
- 1:03:32insufferably polite it is. Like I I think I said this in an
- 1:03:34earlier episode but you know, I beg it to behave like George
- 1:03:37Carlin or something. It it feels like impossible.
- 1:03:40And it's funny, you know, given you're sort of trying to teach
- 1:03:43it to be this like against the system, sort of, it seems like
- 1:03:47person is. It's personality is not disposed
- 1:03:50towards that type of behaviour or how do you how do you address
- 1:03:53that and like why do you think that's happening.
- 1:03:56This is a big issue for us. So in our prompt we definitely
- 1:03:59say you imagine you're an aggressive lawyer.
- 1:04:03The the note that doesn't care about being polite, have to get
- 1:04:07the exact prompt for consumers. They have to.
- 1:04:09They you can put in your own kind of general prompt for the
- 1:04:12entire ChatGPT. And there are some amazing
- 1:04:15examples on Twitter, and some of them are like don't be polite,
- 1:04:19don't apologize, don't say you're an AI, don't caveat it
- 1:04:23with any disclaimers or warnings.
- 1:04:24And that seems to work for people.
- 1:04:26So I would recommend that as well.
- 1:04:28Yeah, I need to put it because I'll put it in the actual text
- 1:04:31in the beginning. It'll still apologize.
- 1:04:33And then I'll say what was the number one thing I told you not
- 1:04:37to do. And it'll, you know, it'll know
- 1:04:39it. I said not to apologize, but
- 1:04:41then it won't implement it. It can be sort of inscrutable.
- 1:04:45I don't know how it prioritises requests it it could be good.
- 1:04:49Maybe they're an open AI program app, because not not being
- 1:04:53polite is a very harmless request.
- 1:04:55But people might want to drag jailbreak it, so they're
- 1:04:57probably trying to build all these barriers to jailbreaking
- 1:05:01it. Yeah, when you write an e-mail
- 1:05:05like, is ChatGPT involved or like in sort of what day-to-day
- 1:05:08tasks like where? How much are you bringing AAI or
- 1:05:12is it mostly around the things your company does where it's
- 1:05:15sort of, you know, yeah, consumer defence.
- 1:05:19I think in kind of communication people can tell.
- 1:05:24I think there's this insult around.
- 1:05:26We heard it during the debates. I think Chris Christie said to
- 1:05:30Vivette you sound just like ChatGPT and I think that's a
- 1:05:33that's an insult and I can definitely tell I have a family
- 1:05:38member. She asked me for my help she's
- 1:05:40writing something important and I and she said I use ChatGPT and
- 1:05:43I said I can tell. So I I try and not use ChatGPT
- 1:05:46when I'm writing an e-mail to people because I think it's so
- 1:05:50inauthentic and and they can definitely tell I do use it in
- 1:05:53other areas of my life. I was playing pickle ball with
- 1:05:55my friends and I asked ChatGPT to create a pickle ball
- 1:05:58tournament for like 12 people and do matches and it does
- 1:06:02things like that very well. But if I was speaking to
- 1:06:05someone, I would never use it at the moment.
- 1:06:08It's good for transactional tasks like negotiating a bill,
- 1:06:11although I'm biased saying that, but it's definitely not good
- 1:06:14sending really high stakes communication path where
- 1:06:19personal touch is needed. In the broadest terms like the
- 1:06:22difference between it being an assistant to an accountant type
- 1:06:26person versus being an accountant, you know, just being
- 1:06:28the agent itself versus supporting people, what's what's
- 1:06:31your general view on how close we are to?
- 1:06:35This is like a total AI problem with maybe a light human
- 1:06:39supervision versus just continuing to help professionals
- 1:06:42who bring in the AI to support, you know, their professional
- 1:06:46work. A few weeks ago I asked ChatGPT
- 1:06:50to add up 10 numbers and it got the answer wrong.
- 1:06:53So. Not close, no, no special.
- 1:06:56Just literally 10 numbers, right?
- 1:06:59It's it's actually not very good at math.
- 1:07:00People don't realize this, but it it's good at language but not
- 1:07:03necessarily math. So I think there's going to be
- 1:07:07software tools where ChatGPT makes the experience amazing,
- 1:07:11like Kit, which is an accounting software for small businesses.
- 1:07:14I know they're using a lot of chat.
- 1:07:15GPTI think accountants will use it, but given that it can't even
- 1:07:20add up 10 numbers on it's own, I wouldn't trust it with my taxes
- 1:07:23yet. Yeah, if it can't add, does that
- 1:07:27mean it really can't reason? You know there is.
- 1:07:30There's an argument that, well, it's just like it knows the end
- 1:07:35you want like get out of this deal with Comcast or whatever
- 1:07:38and it knows what text that might, you know, get it to that
- 1:07:42end looks like. But it it's do you think it has
- 1:07:45like a through line of any sort of reasoning in its strategy?
- 1:07:50When you ask it something, it goes into its model and it says
- 1:07:54based on all of human literature and all these parameters, what's
- 1:07:57the most likely next response, right.
- 1:07:59And so if you ask it to add up numbers, it finds some mass
- 1:08:02training data somewhere and it just thinks, oh OK, well based
- 1:08:06on all the training data, this is the the next response.
- 1:08:09But that's not reasoning, that's just picking picking something
- 1:08:12on a statistical model. And so that's why it can write a
- 1:08:15beautiful poem, but it can't reason to add up 10 numbers.
- 1:08:20Yeah, but do you think it's reasoning in the poem writing or
- 1:08:23it's just good at sort of mimicking in a very
- 1:08:27sophisticated way. It's it's good at picking.
- 1:08:31So it just has a such a big option.
- 1:08:33It can just pick and and throw things together.
- 1:08:36I think it's just really good at picking.
- 1:08:38It's definitely not reasoning at the moment.
- 1:08:41In the accounting context though.
- 1:08:43One, I'll give you a concrete example of why it's useful.
- 1:08:48Previously and we do not pay had this problem, there would be
- 1:08:51like 1000 transactions that you would get from Plaid or manually
- 1:08:55entered and you would it would every transaction would be
- 1:08:59different. There'd be no kind of reason to
- 1:09:02them you can like characterize the transactions.
- 1:09:06So you could say oh PLMP is Planet Fitness or or something
- 1:09:12like that or this is a business expense or this is travel
- 1:09:15expense. So that's where I see it being
- 1:09:17useful in the accounting context, but I don't think it's
- 1:09:19going to be adding numbers, do you?
- 1:09:21Do you do you have any friends in the AI world where it feels
- 1:09:24like I I'm interested in this question of like?
- 1:09:27AI you know just almost like a computer you build at home right
- 1:09:32before you know sort of the mainstream computer is
- 1:09:35available, people sort of cobbling together a super, you
- 1:09:39know a powerful computer of their own based on what they get
- 1:09:43access to. Do you think that sort of is
- 1:09:46happens in a meaningful way or the reality is that just like
- 1:09:49purpose built like Chachi, BT, mid, journey, whatever are the
- 1:09:53best at what they do So people are just using these?
- 1:09:57Obvious tools like do you is there a real like are you seeing
- 1:10:00people get their own memory and blah blah blah to to sort of
- 1:10:04have an advantage over what's publicly available.
- 1:10:08It's not simple enough for consumers to rely on AI just
- 1:10:12with ChatGPT I I agree with your theory that they have to build
- 1:10:15their own computer. It's not even just about the
- 1:10:18AIAI as I mentioned is useless if it's not connected to the
- 1:10:22real world. So how do you connect AI with AP
- 1:10:26is so one thing I do is I connect ChatGPT to the mail
- 1:10:31because it language through mail is one way you can get things
- 1:10:34done and I use the LOB API to do that and so you have to.
- 1:10:38Find a way to old fashion mail. Old fashioned mail, so in the
- 1:10:42legal world you know it's a Stone Age.
- 1:10:45You still have to send mail to get things done.
- 1:10:48Some of these disputes are over mail.
- 1:10:49How do you get chat GPD to send a physical latter and that's
- 1:10:53through the LOB API. Similarly, how do you get
- 1:10:55ChatGPT to make a phone call? Twilio API?
- 1:10:59The building your computer is plugging in all of these janky
- 1:11:02AP is so that the language can get through and then also
- 1:11:07there's this open source component with the gatekeeping
- 1:11:10so it all ties together on it's own.
- 1:11:13Just having an interface where you can send or receive chat is
- 1:11:16not that useful. It only begins to be useful when
- 1:11:18it can be an agent in the real world, and I I think if there is
- 1:11:21one AI law we should pass, we should pass a law that says that
- 1:11:24AI can be an agent on your behalf to help people.
- 1:11:27Fascinating isn't isn't your your dad is like you know has
- 1:11:34that connection to the Russian lawyer.
- 1:11:35Yeah, isn't it? Don't you have a like very pro
- 1:11:38lawyer background or I'm curious your your view on how your view
- 1:11:43on lawyers has developed or where where you see the the role
- 1:11:47for them. I think that there are a lot of
- 1:11:51amazing lawyers out, including the ones that help us and human
- 1:11:54rights lawyers that my family's very involved with like fighting
- 1:11:58back against Russia. But I also think that there are
- 1:12:01lawyers on billboards that charge people a lot of money for
- 1:12:04doing very little and I think that it with AI, hopefully it
- 1:12:08will replace them and make their services free and accessible for
- 1:12:12people. I was in Vegas a few weeks ago
- 1:12:16and there are so many lawyer billboards and I hope AI will
- 1:12:19replace all of them. What just given how much you see
- 1:12:22and like you interact with a lot of these other companies, like
- 1:12:24what, what do you think will happen in the next five years?
- 1:12:28Or like, what's your prediction for how all of this plays out?
- 1:12:32Or, you know, just a couple of things you're most excited
- 1:12:35about. The biggest thing I'm excited
- 1:12:37about is AI going from attractive to proactive.
- 1:12:40So a lot of these assistants like right now you go into
- 1:12:43ChatGPT you say I want this similarly with even do not pay.
- 1:12:47You say I want this my info my wife it didn't work.
- 1:12:51But these a is are going to be integrated into people's systems
- 1:12:54so that you just wake up one day and the AI has done something
- 1:12:57for you overnight. And to get that there's be a
- 1:13:02kind of latency issue so that the AI can do 1000 things to
- 1:13:06check what it wants to do a cost thing.
- 1:13:09It has to be much cheaper because to find out the right
- 1:13:12thing to do you have to throw a lot of darts.
- 1:13:15And then also just a systems integration where a pool will
- 1:13:19have a kind of probably a local LLM on the device constantly
- 1:13:22figuring things out. You'll have lots of data
- 1:13:25integrations with plug in like API type things that connect to
- 1:13:29the AAI. So I'm really excited.
- 1:13:32I'm trying to build a world where people wake up and the AI
- 1:13:35says I've just saved you $100. I think there'll be situations
- 1:13:39where the AI will say maybe you should get your health checked.
- 1:13:42I am noticing something wrong with your blood and and things
- 1:13:45like that and that will really improve quality of life do.
- 1:13:48You think the consumer is on your side or where?
- 1:13:51Where do you think the regular person is in terms of fear
- 1:13:55versus excitement here and how that will translate into
- 1:13:58regulation? Ordinary people love it and the
- 1:14:03the only people that I I think there's a lot of scaremongering
- 1:14:07going on, but I don't think it really touches the ordinary
- 1:14:11person's experience. People just work really hard,
- 1:14:14they don't have much time in the day, they're being ripped off
- 1:14:17across their life. They just need something to help
- 1:14:20them write. Thanks for writing thank you
- 1:14:24notes to saving them money. I think it's it's mainly
- 1:14:27beneficial to them. The kind of intellectualisation
- 1:14:32of of the dangers of AI worries me, and that's probably why GPC
- 1:14:36Four got worse. So we we should start doing
- 1:14:38that. Are you, what do you think of
- 1:14:40the whole like, you know, AI doom and gloom sort of.
- 1:14:44It'll kill us all. Sort of general intelligence
- 1:14:47fears. Like are you do you think we're
- 1:14:49anywhere close or do you have a strong point of view on that
- 1:14:52discussion? I think the, I think there are a
- 1:14:56lot of evil people working on AI right now.
- 1:14:59I've seen this first happen. Typically with a new technology,
- 1:15:02it gets in the hands of evil People 1st and we actually need
- 1:15:06to work on AI, accelerate our work on AI to fight back against
- 1:15:10them. They're going to work on it
- 1:15:11regardless. So I'll I'll give you a concrete
- 1:15:14example. So what we're saying we we help
- 1:15:16people charge back things and fight fraud on their bank.
- 1:15:19We've even seen a few examples of fraudsters.
- 1:15:22They phone up, they pretend to be someone's relative with a
- 1:15:26spoof voice, and you can do that right now with just a minute
- 1:15:29recording. And the way you fight that is
- 1:15:31you have an AI from the Telecut phone company stopping people or
- 1:15:35saying or on your phone maybe device side if you don't want
- 1:15:38the telephone listening. Saying this is not who you think
- 1:15:42it is because our AI is better and we've we've detected it.
- 1:15:46If we stop work on AI, the criminals outside of the
- 1:15:49jurisdiction of the United States will still build their
- 1:15:52technologies, but we just won't have the tools to fight back.
- 1:15:56So it's never a good idea to stop the progress of technology
- 1:15:59because the evil people won't won't stop.
- 1:16:01Yes, I agree with that. The copyright problem I where
- 1:16:05I'm a little more sympathetic. Lots of writers are concerned
- 1:16:08that these these large language models have inhaled all their
- 1:16:12thinking and work to become what they are and that they should be
- 1:16:16compensated in some way for it or what.
- 1:16:18What do you make of sort of the creator and writer class being
- 1:16:21worried about what's been built potentially on their backs?
- 1:16:26I think that there's no practical way of dealing with
- 1:16:31it. No one knows how these large
- 1:16:34language models come up with their answers, and so perhaps it
- 1:16:38can list it by ten sources. But ChatGPT, every answer is
- 1:16:43more than ten sources. It's a million sources or even a
- 1:16:46trillion sources. So I I think if I was to read 10
- 1:16:51books and then write a book myself, even if it was inspired
- 1:16:53by the 10 books, I don't think I should have to compensate the
- 1:16:57people I read. If I physically copy that work,
- 1:17:00I do. But if it's just inspired by I
- 1:17:02don't. I don't think I would have to.
- 1:17:05I I know it's a very tough situation, but I don't see any
- 1:17:09practical way of doing the compensation.
- 1:17:12Fortunately, though, the lawyers have already thought about this,
- 1:17:14and there's like multiple lawsuits, class action lawsuits
- 1:17:17against ChatGPT that will play out.
- 1:17:19I don't want it to be, you know, so costly that it means these
- 1:17:23things can't come into existence.
- 1:17:24But if it turns out, you know. Large language models are fairly
- 1:17:29commoditized. We can you know people are
- 1:17:31competitive with Chachi BT then it would seem like a lot of the
- 1:17:37core sauces like the breadth of human intelligence that went
- 1:17:43into them and if like, I don't know especially specific ones
- 1:17:46where like. Right.
- 1:17:47In the style of like Stephen King or whatever, you know, he
- 1:17:49created an iconic style that's extremely valuable.
- 1:17:53So I I guess I just don't think it's like this large long tail
- 1:17:57for a lot of the stuff. I think it's more likely that
- 1:17:59there are a few core great things that are being used, you
- 1:18:05know for, for a lot of answers and that could be sort of
- 1:18:08attributed and. Those people should probably be
- 1:18:11compensated in some way if for the rest of human history you
- 1:18:14know those those answers are going to be a bedrock of of how
- 1:18:17these systems answer. Do you disagree with that?
- 1:18:21But I I do. I think there's existing laws on
- 1:18:24the books to deal with that. So California has this law about
- 1:18:27right to publicity and likeness, which is completely separate
- 1:18:31from copyright law. And so if I was to say right in
- 1:18:34the style of Stephen King, that would probably violate like his
- 1:18:37right to publicity if I was to publish a book from Chachi PT
- 1:18:41writing in his likeness. So there there are a kind.
- 1:18:45Of but you ask Chachi PT to write a sci-fi novel and it's
- 1:18:48like it's like based on like 5 different, you know, I feel like
- 1:18:52it's based on you know the top ten list of sci-fi authors and
- 1:18:56then it sort of finds the midpoint between them and gives
- 1:18:59you, you know some text. Is that another?
- 1:19:02Another issue is that if I say that that's not copyrighted, I
- 1:19:06don't even own that work. The Copyright Office recently
- 1:19:10made a ruling that things done purely with generative AI do not
- 1:19:14qualify for copyright, even if it's 1000 prompts.
- 1:19:17So if I do a mid journey thing and prompt it 1000 times, fine
- 1:19:20tune it, it's still not mine from a copyright perspective.
- 1:19:24So I I do think these things have to be changed in some way.
- 1:19:28What? Sorry, What you want?
- 1:19:30You want a law change where you get protection if you build
- 1:19:33something out of or? I'm confused what you're saying
- 1:19:35there. If if you fine tune the output
- 1:19:39of a large language model, you're not protected with
- 1:19:42copyright at the moment. So I I do think that the system
- 1:19:45is is not built to deal with the AI and so I'm not arguing that
- 1:19:48we just do nothing. I think there's lots of these
- 1:19:51new answers, but I I do worry that we would just halt
- 1:19:54technology. Yeah, yeah, you're you're you're
- 1:19:57saying, OK, maybe we need to build it, answer some of these
- 1:19:59specific questions, but not not if it's putting roadblocks.
- 1:20:03I'm curious like the. The existing chat bots like the
- 1:20:06characters of the world like are you are you have you played
- 1:20:08around with them Much like do you think they're they have
- 1:20:11perpetual value or what's your what's your view on that space
- 1:20:14at the moment? We're going to have AI
- 1:20:18girlfriends, AI people can AI therapists, AI friends, and
- 1:20:23people can talk to their favorite characters.
- 1:20:26I'm a big fan of the show Better Call Saul for obvious reasons,
- 1:20:29so I was playing around on character AI Better Call Saul.
- 1:20:32But. So I think it's it's really
- 1:20:34important to give people that connection.
- 1:20:36So I'm I'm for it. I don't think it's as big as
- 1:20:40people think it will be there. You think it's overhyped?
- 1:20:45I think that given the kind of modality, it's just text at the
- 1:20:49moment, there's only so much you can do with that.
- 1:20:52So what? Once you have like a slowly
- 1:20:55interactive character that you can FaceTime, and I know that
- 1:20:58those exist, but latency is too big, it doesn't feel completely
- 1:21:02real. Then then it gets serious.
- 1:21:04But until then I don't think people will just spend all their
- 1:21:07time talking to spots chat bots because no, no one likes to do
- 1:21:11that. You sort of open the
- 1:21:12conversation talking about multimodal like clearly that's
- 1:21:16something you're excited about. Where do you think how close is
- 1:21:19that? I mean you know Open AII Guess
- 1:21:21is released something where Dolly and Chachi BT are
- 1:21:25connected, but can you talk about what you think the real
- 1:21:28potential of multimodal is? Yeah, GPT 4 VI think was the
- 1:21:35most exciting thing around that where you can upload images and
- 1:21:38and PDFs and things like that to to make it multi modal with the
- 1:21:42API. If AI is constantly ambient
- 1:21:47listening it gets back to this kind of proactive approach which
- 1:21:51is so exciting chat. GPT is very powerful, but it
- 1:21:55doesn't have the data right now, and so another concrete example
- 1:22:00is the browsing. So right now the way ChatGPT
- 1:22:03browsing works is it takes the HTML of the page ingest that
- 1:22:07produces an answer and that's why it's so time intensive.
- 1:22:10Multimodal fixes that in the future I imagine it takes a
- 1:22:13screenshot of the page and then they can put any blockers or
- 1:22:17anti bot things, it doesn't matter they've taken a
- 1:22:19screenshot. And so how we view things is,
- 1:22:23how humans perceive things is through vision.
- 1:22:25And that's why I think it makes it so powerful because no one
- 1:22:28can stop vision. Sweet.
- 1:22:30Joshua Browder, thank you so much for coming on the show.
- 1:22:33I really this was great. Thanks for having me.
- 1:22:36That's our episode. Do Not Pay CEO Joshua Browder
- 1:22:40Great conversation about the AI personal stack.
- 1:22:44Thanks so much to Max Child and James Woolsterman, the Volley Co
- 1:22:48founders. I'm your host, Eric Newcomer.
- 1:22:51This episode has been in our Cerebral Valley series on the
- 1:22:54Newcomer podcast I'm hosting with Max and James this Cerebral
- 1:22:59Valley Conference in San Francisco on November 15th.
- 1:23:04This is the second AI conference in a year, and you can go back
- 1:23:07and watch the old videos on our YouTube channel from March.
- 1:23:11And now, yeah, we're going to, we're going to have another
- 1:23:14exclusive AI conference, but we're bringing it all to
- 1:23:17everybody on YouTube and some of our favorites will go on the
- 1:23:22podcast feed. Shout out to our producer Scott
- 1:23:25Brody, my Chief of Staff Riley Kinsella, Gabby Caliendo at
- 1:23:30Volley and young Chomsky for the wonderful theme Music.
- 1:23:35Like, comment, subscribe on YouTube, give us a review on
- 1:23:38Apple Podcast, play a Alexa game, try Song quiz or yes, sire
- 1:23:45at Volley. And of course, most important,
- 1:23:48subscribe to the sub stack newcomer.co.
- 1:23:52Thanks so much. Goodbye.
- 1:23:55Goodbye.