gptboss:
Yeah.

Adam:
Yeah, that's, you know, it, it leaps, it leaps very quickly to like UBI conversations around novel ways to say, like, we'll pay you for this. We'll pay you for that. Whatever.

gptboss:
Mm-hmm.

Adam:
This is part of it, but they're not going to be tasks that you can use market value necessarily to, to, to pick the value of, um, like for a lot of those people. Cause even if everybody turns to the blue collar thing that Mike row was talking about, there's only so many of those jobs.

gptboss:
Mm-hmm.

Adam:
Um, anyway, hello to anyone tuning in the live stream. Whoops.

gptboss:
Hello YouTube.

Adam:
Nope. That was only for Steve. I forgot to hit the go live button. Only one minute late now. Hello to anyone.

gptboss:
Hello YouTube.

Adam:
tuning in on the live stream. Okay. What are we doing here? We're doing a podcast.

gptboss:
We got a show.

Adam:
Let

gptboss:
We

Adam:
me

gptboss:
got

Adam:
find

gptboss:
a show

Adam:
the script.

gptboss:
to get on the road.

Adam:
We do have a show to get on the road. Okay. It's Friday, May 25th. This is Accelerate Daily. Coming up today, OpenAI pushes back on regulation. Google will be serving you AI-generated ads soon, Legal Tech AI funding news, and an AI to help you with that novel you've always wanted to write. Also today's request for feature, an AI fact checker. Put your goggles on, let's jump into it. I gotta come up. It's Steve, you can leave this part in, except for the note to Steve. We got to start, we got to start working on that tagline. If anyone's listening, they have a good idea for a better tagline for what we ended up talking about. Uh, hit us with it. Uh, we'll try it out.

gptboss:
Yeah.

Adam:
We certainly have enough days in the year. We keep this project going.

gptboss:
Yeah.

Adam:
Okay. intro music! Welcome back everybody, I'm Adam.

gptboss:
My name is Mackenzie, good morning.

Adam:
And we're back with three links and one recipe to help you stay caught up with what's happening in AI today. All days of the week. How, uh, uh, so we were gonna, so we agreed to do our, our, our first bit here around your, uh, the Tik Tok he sent out earlier this morning, um, which goes with the banter you probably will have heard already in the intro, um, about the tagline and stuff like that. Uh, as a producer of this kind of content, I don't really believe in a lack of authenticity in terms of like owning the artifice that is like, look, we're making a show here. We have an editor. He makes it sound better for us. And then we put it in front of you. Part of that is figuring out what we should even be talking about to be doing a compelling project here. But you had some insights. I thought we're, a good way to kick this one off.

gptboss:
Sure, yeah, so in some on the TikTok, I'll try to move through it quickly because we're doing a show. There's this tension for a content creator. Like if you're making content every single day because you have a mandate or it's a part of your job to do content, one of the, like there's tension in like making sure that there's enough content going out and making sure that the content is good. So what does it mean to have good content and how do you stay on top of this stuff and kind of a default mode is news. hey guys, this thing just happened, here's what that means. So that kind of pattern is great most of the time. However, there's this other larger cultural arc that's been going on where there's competition between news media and other forms of attention, like cat videos on the internet and Magic the Gathering let's plays and so on and so forth. Like the internet has everything that you ever wanted to know and more. So what news media is doing now is becoming and this has been a long, long, long, long time coming, is it's highly sensationalist to capture and maintain your attention because maintaining your attention games the algorithm and it shows it to more people. We have a more successful and better show if we're showing you things that scare you. If we show you things that you hate and you're like, I don't like this at all and I'm commenting to let you know that I didn't like this, that is technically a better show. than

Adam:
To

gptboss:
something

Adam:
the algorithm,

gptboss:
that you liked

Adam:
right?

gptboss:
and you didn't

Adam:
It's more

gptboss:
comment

Adam:
engaged.

gptboss:
about.

Adam:
It causes

gptboss:
Yeah.

Adam:
you to watch more ads. It's

gptboss:
Yeah.

Adam:
what what YouTube wants you to yeah It's it's a part of it, right? Which is which gets to like hey if you're watching this on YouTube If you're listening to this on YouTube Right now I know from the analytics that you probably found us through the recommended thing I have no idea what video fed into it, right? But like the Stuxnet one, you know that that headline reads reads like Like, like scary. Yeah.

gptboss:
That's

Adam:
Fudd

gptboss:
a threat, right?

Adam:
fear, uncertainty,

gptboss:
Yeah.

Adam:
and doubt.

gptboss:
Yeah.

Adam:
Um,

gptboss:
And

Adam:
but

gptboss:
that's

Adam:
it's

gptboss:
our

Adam:
our

gptboss:
most

Adam:
most

gptboss:
successful.

Adam:
performant video so far. Exactly. So, you know, immediately it starts that question of like, okay, we don't, we're not here to lean on that kind of thing. We hope anyone who found us because the headline read that way or something, because it's unavoidable when you cover the news. Um, but you're right. You know, it's an interesting project to try to hit the news every day.

gptboss:
Mm-hmm.

Adam:
Um, I, I, I'm inclined to throw this conversation out so early because. It's it's more than that when you're doing this at the startup level, where we are right now, which is, yeah, we're talking about those things. We have a mandate to make daily content partially because as marketers, that creates the best data for us figuring out how to, how to solve the problem that we're trying to solve, you know, as entrepreneurs over here, uh, but it comes down to community. And the thing of like, if you want to hear about other stuff, let us know. Get in the comments.

gptboss:
Yeah. So

Adam:
Uh, right.

gptboss:
I'm partially

Adam:
But.

gptboss:
like rebalancing or recalibrating like who this content is for, right? Like

Adam:
Yeah.

gptboss:
we had this kind of conversation about maybe the best people for this because like we're both platform companies. I'm not a part of mission control, right? They don't even pay me to be here. So there's no like we just we both are two separate companies that offer platform solutions for achieving goals with

Adam:
Complimentary.

gptboss:
AI. So,

Adam:
Complimentary

gptboss:
yeah.

Adam:
platform.

gptboss:
Not competing, complimentary, because we

Adam:
Right.

gptboss:
serve different segments of this same audience, which is product builders, right? Like somebody who's in charge of some offering in the market. And so I don't know how interested those people are in daily news. And another thing that you'll see throughout this show, I predict, is that some of the things that we bring up are going to be very cool, but we're not going to be able to bring it home for a product owner. and say, this is how you can start using this today, because most of the cool things that we see on the horizon are waitlisted. So we can't give you advice, we can't go, here's how I use this to automate my things, and you could do the same. You know?

Adam:
Right. Well, and so that's why it gets to the conversation of, you know, kind of what we're doing here, right. Um, get in the comments, follow the links, tell us how you found us. If you're interested in keeping up with this stuff, we can do a half hour a day on just cool ways to use AI. If that's what people really want to hear that have, you know, clicked on these things, uh, listening to the podcast, whatever they're doing. Um, But for now, still hitting the news, getting some stuff at the end of school. But also we know that people are watching. So

gptboss:
It's

Adam:
thanks

gptboss:
not like

Adam:
for,

gptboss:
cool stuff didn't happen.

Adam:
thanks for stopping by. Yeah. Um, but I think the main message is like. If you're here, stick around to be stoked about the acceleration of this stuff because yeah, there's scary stuff and we want to talk about it because we're professionals here and we have to look at the pragmatic side of this stuff rolling out, but also I work in this space because it's cool as shit and there's all kinds of fun stuff happening and that's really what we're here to talk about. Also, if you know of fun stuff, let us know. We'd be happy to cover it. It would make my job easier. Finding the stuff every morning. Okay.

gptboss:
So we spent a little bit of extra time on that. Let's zoom through this,

Adam:
That's fine.

gptboss:
the title card. This is without question, the stupidest looking animal in the universe. And it could have gone stupider, but there's some weirdness going on here. So it's got eyes, it's like a toad that's very fat and like ugly and hairy, but it has predator eyes. Did you notice that, Kurt? The eyes

Adam:
Yeah.

gptboss:
in the front give binocular vision so it can.

Adam:
Right. The yeah, it's just kind of like a lump of

gptboss:
Hehehehe

Adam:
a, of a job of the hut situation, except a stupid non-threatening job

gptboss:
Hehehe

Adam:
of the hut, like sitting on a log with a galaxy in the background. I have no idea how this is just.

gptboss:
without

Adam:
This is

gptboss:
question,

Adam:
almost

gptboss:
the

Adam:
the

gptboss:
stupidest

Adam:
plug

gptboss:
looking

Adam:
for

gptboss:
animal.

Adam:
our slash mid journey more than it's like, Hey, here's a cool AI image. Uh, it's just hilarious. It's

gptboss:
Yeah.

Adam:
good Friday. Good Friday vibes. This is a stupid

gptboss:
I

Adam:
picture.

gptboss:
wonder what it eats if it is the stupidest thing. What could

Adam:
Yeah.

gptboss:
it be predating?

Adam:
But also, like you said, predator eyes, so it's it's going to be best equipped to just like fall forward on any

gptboss:
Hehehehehehehehehehe

Adam:
passing prey or something. I don't know. Oh, OK. Let's do this thing. First up, we got open AI pushing back on EU regulation from Reuters. Open AI may leave the EU if regulations bite says the CEO, which is Samuel. Uh, yeah, throw this one in mainly as a contrast of the stuff that he said to Congress in the U S and, and I think it's, it's actually sort of to counter what I think is the easy narrative here for a lot of people covering it, which is like, Oh, what a hypocrite. Like, no, that's the dance right now. There is a government trying to be really aggressive over in one corner and a company, the leader of a company has got to go and say, Whoa, whoa, whoa. That's that's. That's hard enough that it doesn't make sense for me to do business in your country. Like it's not all a chess game of jerks. Right. Um, On the other side, he's asking for regulation out of the government of his home country. Cause at the same time, he believes that there should be laws about this stuff. These things can go co-exist. It's the dance of how we get to, to, you know, functional regulation.

gptboss:
I read the draft from the EU. Did you read it yet?

Adam:
Yes, it's, it is. I agree.

gptboss:
A lot.

Adam:
It's very aggressive. There's

gptboss:
It's

Adam:
some

gptboss:
a lot.

Adam:
stuff in there that as a business person, I go, Ooh, that's gonna, I don't know that like as a business person and technologist, I. Look at it. And I go, Oh, that's like, there's a reason the 230 carve out exists, right? Like there's a thing social media caused to have happen where we went, you know, if they have to be liable for every one of their posts, social media companies are going to go under under the weight of. users on their platform who say ridiculous things. It's a completely like it exists for a reason, not just because of politics. Um, we need to start figuring that stuff out around, around AI, but yes, Europe's was very aggressive and there are for sure things in there that I look at and go, Ooh,

gptboss:
It's

Adam:
maybe,

gptboss:
so long

Adam:
maybe

gptboss:
and

Adam:
we

gptboss:
it's

Adam:
just

gptboss:
so

Adam:
don't

gptboss:
complicated.

Adam:
run our company in the, in the

gptboss:
Yeah.

Adam:
U.

gptboss:
It's well, I don't know if you guys are doing found it's like, it's mostly religious station for like foundational models, right? Um, but there's confusion on like what, what a foundational model even is from

Adam:
Right.

gptboss:
news media reading that. Like I, so I got this from a client request. I told them that I hosted the code for their client work on GitHub and they're like, Oh, isn't GitHub AI like illegal in the EU now? And they sent me this legislation. because they saw in a news article that GitHub would be liable for AI things. But it was talking about open source foundational models. They didn't want just anyone having access to like Vicuno or Lama or MPT. Right. They want to make sure that these open source things have a minimum requirement of transparency so that the EU can know what the hell the model is actually doing. Which is questionable. I don't even know if the Vicuno maker knows what it's actually doing. Right. And so the specific quote, the thing that's highlighted in this Reuters article. is that OpenAI would have to disclose what copyrighted material, if any, was used to train it. And they might not know, right? These foundational

Adam:
Right.

gptboss:
models are so big with so much data that they might not be able to comply with some of these seemingly reasonable transparency requests.

Adam:
Yeah. And so, so some of that is the regulation can't be that crazy. Like he has the right to say this, if it's too difficult, they need to back away from some of that stuff, align it to where we are with technology. I think the weird part about it is it's just like, it's the beginning of explosive takeoff potentially, like we don't know the acceleration curve. You can fight about that part in terms of how, how much we're accelerating. Right. Um, But you can already see from the discourse that we're ahead of like the human powered mechanisms for figuring out how to regulate this stuff in terms of he's going, okay. But then I can't offer this service here because I can't meet a couple of these promises they can retrain compliant data sets like over time, but is the world ready to wait a couple of years while we sort it out and they go, no, you can't have this thing anymore or the business leaders of the world ready. For that to not exist, that's the real, you know, the money wants this.

gptboss:
Yeah, yeah,

Adam:
It's very clear.

gptboss:
about half of my customers are in that like geographical

Adam:
Yeah.

gptboss:
region. So this would be like a major blow to me

Adam:
Yeah.

gptboss:
if I can't serve it there anymore. So I hope they figure it out. But speaking of acceleration and waiting,

Adam:
They will.

gptboss:
our next story is a little bit of a tortoise in the hair moment.

Adam:
For sure. It says it's. Generative AI Google ads. Read headline, that's a, I wrote a headline for our card in here that's a mouthful. Anyway, from Wired. Google will soon show you AI generated ads. Advertisers can save money and perhaps make ads more clickable using chat GPT style tools to generate text and images. Meta is testing similar technology.

gptboss:
The images are really good, you guys. Like we've all used Mid Journey and we've all tried to use Mid Journey to write like an ad and it doesn't come out quite right, but these are very specialized foundational models specifically for writing ads. Kirk, I can't read the story without subscribing. Damn.

Adam:
Sorry, I forgot I was logged into a subscription when I read it.

gptboss:
So they're specialized for like, this is where the product is and this is the text on the product, which a lot of. generative images struggle with. They really, really struggle with text. But Google has solved that. So what's exciting about this one is, for me, I've kind of been like, well, obviously, Google has the resources and the compute space to be doing really well with AI. But where are their products? Where is the Google AI thing? Where is your stuff, Google? And so they've been waiting. They've been building. And now I'm starting to understand why. Because this capability of having the same image across multiple generation sets, like the exact same product across multiple sets, masking, proper text, and then text prompts for all of it is something that not even like stable diffusion is capable of right now. So this is, when this releases, this will be the best generative image model available. And so that's like the tortoise and the hare story here. Like they're doing stuff that we kind of don't know about to get to results that we currently don't have access to. And that's really exciting.

Adam:
Yeah, from the beginning, the narrative about Google, how did Google lose the race? Uh, no, they just,

gptboss:
They haven't started.

Adam:
they've been working on the project for a while.

gptboss:
Yeah.

Adam:
It may have been backburner, right? You might go, I've only been a team of 10 scientists for a little while. Yeah. Well, look what you can do with chat GPT. If those 10 scientists already had it for a few years, which is worth thinking might be the case from the public models, maybe not. But they also control this whole other thing, which is the search ads and all that kind of stuff, right? Like what I thought was the coolest in this demo as somebody working at a tiny startup as a marketing leader, although I was at a big company before that. Like, uh, one of my favorite tactics to use in a place where I am right now, but also that I was stunned to find they didn't do in these bigger companies. Still. It's like. just content testing at the top of the funnel. Like it's an okay test to run an ad just to see if that message works. And also if designs work. And so all of our ad campaigns started with 25 pictures. And then we

gptboss:
Yeah.

Adam:
would slowly winnow it down based on like, hey, this one is converting and we didn't expect it. You know, Cafe Nerd did really well for us in a campaign and we didn't, we just picked that randomly.

gptboss:
You never know the thing that's

Adam:
Yeah,

gptboss:
going

Adam:
you

gptboss:
to work,

Adam:
just

gptboss:
right?

Adam:
don't,

gptboss:
So

Adam:
it happens all the time in marketing.

gptboss:
yeah, yeah.

Adam:
And so.

gptboss:
Split

Adam:
One

gptboss:
testing

Adam:
thing that.

gptboss:
is more than okay. It's vital. You have to do it.

Adam:
Right. So,

gptboss:
You have to.

Adam:
so, uh, so we, so there's a lot of tactics that we're always like, they're like mice, you know, it's part of the secret sauce of understanding this stuff. If you're a marketer that gets hired onto things and you go, okay, I'm going to whip your ads into shape. Part of it is do things like this. A lot of that has gotten easier over time. I've been doing it my entire career and it used to be entirely manual. Now it's evolved into like, you come up with all the headlines and you put them all in and then they run different combinations of the stuff,

gptboss:
Mm-hmm.

Adam:
and then you start seeing certain ads do better.

gptboss:
Yeah.

Adam:
And you can do the same with creative, but creative has

gptboss:
Yeah.

Adam:
always still been upload a picture,

gptboss:
Yeah.

Adam:
right? And then you, and you do your own testing around the picture configurations and stuff. So when you see that that was already happening, this piece is pretty inevitable,

gptboss:
Yeah,

Adam:
which is

gptboss:
totally.

Adam:
to say, look, we also know which pictures perform the best across an amazing range of data, this thing can make for you, for you, pictures that you may not have learned from testing or experience actually are the ones that convert.

gptboss:
Yeah. After being nice to Google, I would like to plug, you still have to wait for this, but the text portion is available on gptboss.com. I recently added a feature where you could put your landing page into the chat. It'll read your offer. And then if you put instruction, Hey, can you come up with 10 Google ads for this link? It'll go out, read that link and then come back to you with those 10 ads. And then you could say, make 20 more, make 30 more. Right? So if you want to start running these split tests with eight, let's chat GPT style tool. to write ads. This is something that I've been working on since January. It's May, so we're at month five. It's been in public, I've been getting public feedback and people are using this to drive a lot of results already. So if the images I don't have, I'm sorry, but if you wanna start using AI for writing your ads and running split tests, that's available right now at gptboss.com.

Adam:
Which like you said, this isn't live yet. We don't know what it'll look like once it pops out. Um, if it'll cost any differently than, than other things. And also you got to consider what you're using it for, right? Almost all of their demo content was like product ads, you know, like, like can we put a, you know, a saying on a mug and then put that mug in a thousand settings, um, that's not going to work the same if you're selling a law firm.

gptboss:
And y'all at home, cross your fingers for me that the image will have an API that I could integrate so you could get the images on GPT boss too.

Adam:
Okay,

gptboss:
Moving on.

Adam:
moving on to to to. Legal tech news. Yeah, you know, who knows based on who is listening or watching this thing, but I don't think I've brought it up before. I am technically by training a lawyer. My previous startup was a legal tech company. So I still watch this stuff. And AI is eating that space as well. From Semaphore, the headline says, money pours into AI startups that save countless hours of legal paperwork. Uh, this is largely though, about a friend from legal tech communities, uh, website, rally legal. They were called for a long time document management stuff, but there's all they've always been at the edge of how you deal with your documents at a law firm. So they've got a, you know, some users, not users, enough users to raise a bunch of money for that tool and have recently been building something called spell book, which uses AI to do legal contract things. Um, Super exciting for your lawyer. Check it out. Um, as a fellow entrepreneur, I agree that this is a place that should be eaten by AI. So lawyers can't charge you so much money to, to defend your rights under the law.

gptboss:
Yeah. Yeah. I hope that Laurion gets to a place where it's about like the creativity of argument instead of like who can remember Keisle the best.

Adam:
Yeah, right.

gptboss:
Yeah.

Adam:
But congrats to them raising some money and, uh, personally stoked that AI is eating the legal profession. I think most people would be, if you look at the sentiment analysis on when it comes to lawyers as a profession. Okay. Next up, uh, here's, here's a creative one to wrap this up on, on, on your Friday AI novel writing. Uh, from the verge, the headline says I tried the AI novel writing tool. Everyone hates and it's better than I expected. Um, this is kind of a long read for the weekend, but the author runs through the entirety of trying to write, uh, like a sci-fi novel with one of these tools. Um, and it's not bad. And this, and the, and the thing that it makes me think of an early experience I had with LLMs. That was for me, a moment of like, Ooh, okay. This really has progressed since the last time I was playing with it. Um, and it was asking it to make a joke involving, I don't remember. It was, it was actually a prompt I got during like a talk or something that was just tell a joke about something and something and something, and it wrote a joke that really was funny, um, in, in ways that a comedian would look at and go, Ooh, that's an important subtlety that makes that a multi-layered joke. Um, but if you ask it to do it a bunch of times, it only does that one out of a certain number of times, so there's still sort of like a probabilistic aspect to getting that output. So like, it doesn't surprise me that a project like this is able to have pull quotes that as I was reading it caused me to go, Ooh, that's a good line. But this is a crazy one, because like how long it takes to write a novel? Really long time. How do you feel about this one is I'm curious. So I'm curious Mac in terms of, you know, I know you have tools that can help people with exactly this sort of thing at different volumes.

gptboss:
There's

Adam:
Like what's the, what's the challenge here? Right? Like what's the probability that this thing you get in that the average person gets in there and actually has the experience that this author is talking about.

gptboss:
The hardest part about a novel is no longer like the writing, I think. Like, so

Adam:
Right.

gptboss:
with the question prompt being, do you think that an average person could write a book with this? Yes. Do I think that they could write a good book? I think that storytelling is a completely separate skill from writing and editing. And it's been awkward. Like part of the huge difficulty of writing a novel is that you have to get this artifice, this technical ability of like... Here's how words go together. Here's how you describe sight, smell, sound, data. And then fold that in with a story that is, like it has engaging enough characters and they have enough development, as well as the thematic events of your story is something that resonates with people that they enjoy and they feel like they've learned something and they would recommend this book to somebody else. So those are the next skills to work on. I think if you already have those skills, or if you don't care about not having those skills, this would be a really good tool. And then moving back to like the market, which was my first instinct on this. There is a lot of tension between like all in one, right? Like my AI friend, like my, like limit for my Dunbar's number of like what tools I use. Um, you kind of want to have your AI friend like do the most that you possibly can, but for specific tasks, you're always going to be better served by specialized software. So this tool, um, the tool mentioned is a pseudo, right? and it looks really phenomenal for the specific task of fiction writing. So if you're interested in fiction definitely try it out. I'm going to let the Discord know about this. I'm just seeing this for the first time today. Just hearing about it now. It looks cool. Looks really cool.

Adam:
Yeah, it's a... think it's easy to look at this and think about the job loss piece, because there's certainly a class of creative who has, you know, cut their teeth on you know, cut their teeth to get where they are, but like their whole value prop is I can do the part of putting this into words better than other people. Um, which like you said, it's almost like the technical part. I think people would argue there's an art to it as well, but That's, you know,

gptboss:
It's like

Adam:
is

gptboss:
blacksmithing,

Adam:
what it is, right?

gptboss:
you know what I

Adam:
It's

gptboss:
mean?

Adam:
a

gptboss:
Like

Adam:
thing.

gptboss:
it's equal parts

Adam:
Yeah,

gptboss:
like

Adam:
exactly.

gptboss:
art and technical ability. Yeah.

Adam:
Yeah, absolutely. Uh, this, this minimizes the technical ability. And to me that the exciting part there is like, you know, how many. Dad's making up, you know, bedtime stories for their kids out there. have have what it takes to tell that story and make up the thing, but can't draw or, or, you know, don't. They're not good with a turn of phrase in such a way as to, for the world to never see that,

gptboss:
Yeah,

Adam:
um,

gptboss:
we'll see that idea.

Adam:
you know, and, and if we

gptboss:
That's.

Adam:
then assume that markets will continue to reward whatever the best stuff is, then we're going to have some like kids stories, books, like whatever version of sound cloud. Where you get little Nas X's just out of nowhere and go,

gptboss:
Yeah,

Adam:
wow,

gptboss:
yeah,

Adam:
that's, that was the whole thing,

gptboss:
yeah,

Adam:
right?

gptboss:
totally.

Adam:
Um,

gptboss:
Yeah, I'm really

Adam:
uh,

gptboss:
excited for that.

Adam:
uh, uh, Justin Bieber, you know? Um, and I guess you could have your qualms with the quality of those things. Right. But like, it's a pop culture algorithm surfacing stuff. People like.

gptboss:
Mm-hmm. It's interesting that this is happening. Long form content is getting easier to write at the same time that attention spans are diminishing.

Adam:
Right. So that part of it will be interesting because it's all a war for attention. Like we talked about at the top of the episode. Okay. Let's get to the recipe for today.

gptboss:
You know what, Kurt? I saw an ad for this exact product right before the

Adam:
This

gptboss:
show.

Adam:
exact thing.

gptboss:
This exact thing. Literally this exact thing. Do you want me to send

Adam:
Well,

gptboss:
it to you?

Adam:
we should look up what it is and we can send a link or something,

gptboss:
Yeah.

Adam:
but this was meant to just be sort of a, okay, well, like what would the challenges of actually doing this be awesome if it already exists, because it's a thing that we need, but I can also see where that thing might be, you know, like now it becomes a question of how do you audit that thing to have it be doing this properly, but the request for feature is fact check AI. Um, the simple flow chart here is fact. to an LLM, to a real time score. Like how close are we to having a trustworthy oracle that can run during a Donald Trump town hall and effectively fact check all of the lies? Like faster

gptboss:
Oh

Adam:
than human moderator can because the human moderator only gets to speak up every once in a while and there's all kinds of politics around what they're allowed to say, like the problem

gptboss:
Mm-hmm.

Adam:
with, and I single out Trump because he's an epic example, but like it's, it's rampant

gptboss:
Oh, haha.

Adam:
all over the place that we have a problem of like, I don't know what to trust and, and, and people are just saying what they believe like

gptboss:
Yeah.

Adam:
as fact

gptboss:
You

Adam:
and.

gptboss:
know, it would go crazy if it was like hooked up to like an alarm. So like somebody's like on stage saying something and then they like finish

Adam:
Hahaha

gptboss:
their sentence. And there's just an extremely loud, incorrect buzzer noise.

Adam:
But so in terms of the technical challenge,

gptboss:
That is not what the product is, by the way. The product I was looking

Adam:
yeah.

gptboss:
at is like, you write and then you make a claim in your writing and then it searches the internet to find a source for that claim and

Adam:
Okay.

gptboss:
then links to it.

Adam:
Yeah. No, my idea was more like, can I get a browser extension that fact checks news sites that sounds like probably based on what you were just saying? Um, and then the more aggressive

gptboss:
I've had...

Adam:
version would be the thing I just pitched about like political debates. Right.

gptboss:
Yeah, I had a, um, uh, person from Tik TOK reach out and say, Hey, this is my product idea. I want to have a fact checker extension. And I said, I don't want to go near that. I don't want to do that. I don't want to be that guy because that is an Oracle task. Right? Like if you're on a new site and then you want to say that fact is wrong, then the only way that you could do that is by referencing an Oracle that you either own or approve of. So this Oracle problem is, we're super used to Oracle problems from crypto to try to get like off-chain activity on-chain. But in the case of AI, it's like single store of truth, right? So how do we have a democratically controlled single store of truth that is never wrong? I don't know if that's a possible task. I don't know if that can be done. And that's what this requires.

Adam:
So that, so I was prepared for that. Uh, do, uh, but I forgot to throw it out early on as part of the premise, ignoring the Oracle problem,

gptboss:
Hahaha

Adam:
which is that, okay, this idea is going to break down on some level of, yeah, but there's possible bias in the Oracles you've chosen for your whatever, right? I think a more, a cleaner way to look at it where you can see that it is maybe a technical challenge that is solvable. except for where it hits human, arguing, meet computers with feelings.

gptboss:
Mm-hmm.

Adam:
Is like the, the provenance stuff we were talking about with images. Um,

gptboss:
Mm-hmm.

Adam:
you can imagine a consortium of camera makers agreeing to put a tag into all of their systems that will apply this cryptographic key that lets anyone check through the provenance of this stuff and Adobe and all the image makers are on board and all the professionals are blah, blah, blah, and, and, you know, a publication can reliably know if the image they're using is, you know, has whatever rights attached for intellectual property purposes. Technically solvable people working on it.

gptboss:
Another

Adam:
That

gptboss:
way?

Adam:
part of that part of it aside, my question is, is AI to a point where it can conceivably like take the ideas from that thing, parse it, be some sort of registry that, cause we could also approach this from a understandability and transparency standpoint, which is to say like foundational models that do X. Which is fact check against a giant corpus of, you know, like compliant journalism, basically. Uh, We could build that, right? I mean, you can train that model now that can consume that the, the input of texts or transcript or audio, and actually flag, which things are not true if you can solve the or. Crazy. Ha ha ha.

gptboss:
Yeah, I don't know. So another way to deal with it, I'm reminded of raging against the machine, right? I have Testify going through my head.

Adam:
Yeah.

gptboss:
Who controls the past controls the future, right? So that Oracle problem is something that's so pernicious that I really want to avoid. One way to deal with it is the way that Twitter does it, which is community feedback. But it does not work for emerging news. You know what I mean? There could be. There could be. some database like this, but it's unlikely that there would be community feedback on enough data points that every single talking point in a political debate would be covered by this database that is democratically controlled. Very, very, very difficult problem. But wouldn't

Adam:
It's

gptboss:
it

Adam:
interesting

gptboss:
be nice?

Adam:
though, because what you talk about is functionally direct democracy, right?

gptboss:
Oh,

Adam:
You

gptboss:
I

Adam:
got

gptboss:
love

Adam:
to be involved

gptboss:
direct democracy.

Adam:
if you want the public score, if you want the public square to not suck and it's like, okay,

gptboss:
I

Adam:
but

gptboss:
promised

Adam:
only

gptboss:
I

Adam:
the

gptboss:
would

Adam:
most.

gptboss:
never make policy recommendations on this show, so I'll just say I like direct democracy. I promise.

Adam:
Um, I do too. Uh, but yeah, I think that community point is, uh, a great, the great, a great, a great pivot point for us to get out of here. Um, community feedback, like you just said, like we said at the top of this episode, if you made it this far, let us know what you think we review, uh, subscribe, don't subscribe, get in the comments. I don't know. We're publishing this a bunch of different places. Uh, but let us know what you think of this problem, how we might solve this problem. Right? Like this is a very real thing in a world where the, the, the people who are trying to tell lies are going to be able to flood the zone even harder. Word it's Friday and it's a long weekend. So we won't be here Monday, but we'll be back Tuesday for more accelerate daily. I've been Adam.

gptboss:
All right, I've been Mackenzie, see y'all.

Adam:
Take it easy everybody.