Kerp:
It's June 1st, this is Accelerate Daily. Today we've got Japan's moves on copyright and what they mean for the intellectual property space as an AI battleground. OpenAI talks about their future plans and GPU constraints. And yeah, I guess the end of humanity if we're tracking with the cover of Time Magazine. Also a prompt workshop on using generative AI tools to summarize things. Get your hand away from the... Get your hand away from that abort button. It's time for Accelerate Daily. That wasn't the best read, but I'll get it eventually. So far the Abort button is my favorite.

gptboss:
step away from the plug.

Kerp:
Turn it off and back on again. Okay. Welcome back everybody. I'm Adam.

gptboss:
My name is Mackenzie. Good morning.

Kerp:
And we're back with three headlines and one how to, to keep you caught up on what's happening in AI today. Uh,

gptboss:
Yes,

Kerp:
Mack,

gptboss:
sir.

Kerp:
what are we looking at for today's AI image of the day?

gptboss:
This is a hippie that's like super crazy. But to me, he's just a regular hippie.

Kerp:
Do you follow the links before

gptboss:
I do.

Kerp:
this?

gptboss:
Yeah.

Kerp:
Okay. Uh, this is the Joker as a hippie.

gptboss:
Yeah.

Kerp:
Um, But yeah, not far off a regular hippie,

Kerp:
except with that Joker vibe. Like I feel like you're up to no good. Anyway, part of a series, um, containing a bunch of DC superheroes. Uh, there was no prompt for this one, but the title shared on Reddit was Batman, Joker, and other DC characters as hippies.

gptboss:
pretty brutal.

Kerp:
Check it out.

gptboss:
Yeah, the

Kerp:
The flash is funny.

gptboss:
The flash, yeah. Or, yeah, Thor just looks, oh, Captain America. Like a lot of them just look like dead heads, right?

Kerp:
Yeah.

gptboss:
This is wicked, I would hang out with these guys.

Kerp:
10 out of 10 would hang

gptboss:
Yeah.

Kerp:
out. Huh. Okay. Uh, so we, before we jump into the topics, a reminder to like, and subscribe, wherever you're watching, listening, throwing a comment, write a review. These, these all help us reach the right audience and the algorithms and feeds so we can keep answering the questions that community brings to the table. Okay, let's jump into the topics. else will work. Oh yeah, out of the gate with the heavy one. Um, this is an example. Okay. Uh, is AI an arms race? This is, uh, the cover of time magazine that says the end of humanity with AI and humanity highlighted. So the other link is to time magazines. The, the lead article AI is not an arms race. Which really isn't as scary as everything I

gptboss:
Hehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehe

Kerp:
had to click to find it. Um, which is a little bit just the state of journalism, but also like it shows the Overton window on the conversation, which is the thing. Like we like to call out over here. Cause just six months ago, we were having completely different conversations with compliance officers at companies, um, as well. which is like, uh, it's bananas to watch just the explosion of like, Oh yeah, this is maybe a problem. Uh, but the article itself is about the idea of whether or not, um, This is an arms race or like something different. Basically.

gptboss:
This is my second time appearing on the cover of Time magazine. The first, of course, being 2006 Person of the Year.

Kerp:
Ha ha ha.

gptboss:
And it's pretty cool. So my takeaway from this article is I'm pretty sure this was written by like a PR company because it aligns so well with the mission or the statement that we were talking about yesterday from safe AI, which is just like identifying AI as a threat. They're not actually talking about what the threats are in the article at all. There's no concrete examples. Um, they're just like scary words, scary words, scary words. And so, you know, Kind of the metaphor at the end of the day is that we as humanity are all standing on a very thin pond, frozen pond, right? Thin ice. And there's a pile of treasure on the shore. And if we all move slowly together, we'll get to the treasure. But if somebody sprints ahead, they might break the ice and doom us all.

Kerp:
And so,

gptboss:
LOL says the sprinter, LMAO.

Kerp:
right.

gptboss:
Right? Like, there's no, how do you disincentivize? It's an infinite prisoner's dilemma. How do you, without communicating with the

Kerp:
Yeah.

gptboss:
other prisoner, disincentivize their behavior? And the

Kerp:
The

gptboss:
answer is you can't.

Kerp:
yeah, I mean the reality on this one and and and and maybe this diff not maybe this for sure differs from Like a nuclear arms race because it's really hard to make a nuclear weapon like the side the technical engineering of building the thing and having it actually work a feat of engineering difficult to do. Digital stuff is just oops, I copied and pasted a bunch of secrets into a discord. You know, um, so. Thing I wrote, you know, it definitely feels like the hype, the, the, Hey, we need to worry about this sector of the hype machine is running. Um, the note I wrote as I was reading this was like. There was a point, there was a thought provoking point in there for me, which was that, uh, this isn't a race because it should be a collaboration with the realization that there's a third agent, right? Like you, you nuclear proliferation is up against physics. Uh, this is up against this agent that we're talking about coming online. That's way super smarter than us. That has its own potential incentives, creating negative pressure against whatever, uh, or, you know, influence, I should say, I want to call it negative pressure, like in one direction or another, it's like if nuclear energy could go, yeah, could, could actually say to you, yeah, but nuclear energy is good. That's good. And everyone goes, okay. It's so in that sense. It is a weird different thing.

gptboss:
Yeah. Is

Kerp:
The

gptboss:
it

Kerp:
end

gptboss:
like?

Kerp:
of humanity, it's like all technology. It makes humanity different in some kind of way.

gptboss:
There's something very like Landian about it, right? Like it's like that argument of like, there's a third agent is like, oh, there's something that's like invading us from the future that doesn't exist yet. Replete with sinophobia. They specifically call out a Chinese lab for not caring enough about ethics or safety. And that would be the person sprinting ahead.

Kerp:
Yeah, it's, it's, uh, that's the part of it. That's hard, right? Cause they're right there. It's like, it runs into the definition of intelligence problem, right? Can you hold two, two opposite thoughts in your head at the same time? Uh, both things are happening. There's a race condition. It's going to be based on open source software, blah, blah, blah. And you could try for enforcement, but it's going to look like trying to shut down Napster. Um, and then there's. The side of like, do you need to have a license to continue to develop something like this or whatever?

gptboss:
And that's, that's why this feels like a PR thing, because they don't

Kerp:
Yeah.

gptboss:
identify any specific dangers. They just say, this is dangerous. Again, like repeating that same docking point,

Kerp:
Right.

gptboss:
uh, talking point from safe.ai, which didn't identify

Kerp:
Thanks for

gptboss:
any

Kerp:
watching!

gptboss:
dangers. And a lot of people have questions. Um, I, cause I posted that story to tick tock and a lot of people were like, why would an AI want to harm us? That's

Kerp:
Right.

gptboss:
not being talked about in this part of the hype cycle. Nobody's identifying what the actual danger is. They're just like, Oh, it's smarter than you. So watch out. That doesn't. It's not, you know, what I mean, like there's, they're, it's almost like they're so not sure, or they so don't believe what they're saying about this being dangerous, that they can't point to anything concrete that would back up their claims. It's almost like that. It's not quite like that. I think

Kerp:
Right.

gptboss:
that they do believe this, but it has that vibe too, of like, just sources trust me, bro, right?

Kerp:
It's trust me, bro's all the way, all the way down. Unfortunately. Okay. Moving on. This one is thematically linked. Copyright as a battleground, uh, from technomancers.ai Japan goes all in copyright doesn't apply to AI training. Um, this was a, this was a doozy to verify cause you had to click through to like Japanese ministry of something and it was not in English. Translate buttons are magical. But there's a summary right here. Anyway, they've come out and said that, uh, that it's just like, there's no copyright protection with regards to, um, using intellectual property in your training data

gptboss:
Mm-hmm.

Kerp:
sets. Uh, there's another case in the U S going on right now, Getty versus. I think stability around over over stable diffusion. which feels

gptboss:
because

Kerp:
like it might play out differently, but this is an example of the aforementioned race condition problem, right? Because if I wanna sell albums in Japan. or, or I want to publish MP3s in Japan, I have to realize that they're going to get scooped into an algorithm where a thing might, you know, whatever, right? Also Japan is generally savvy on this kind of thing in terms of winning a wave of technology, uh,

gptboss:
Yeah,

Kerp:
enabled

gptboss:
yeah, common

Kerp:
commerce.

gptboss:
Japanese W. They do this all the time.

Kerp:
Yeah.

gptboss:
One of my favorite Japan jokes is like, you're in in the Yamaha store, and you're like, this piano is great. But do you know where I could get a chainsaw and a motorcycle? And

Kerp:
Hahaha

gptboss:
they just, yeah, Japan's wicked. This is I think the correct take personally.

Kerp:
Yeah.

gptboss:
Because like, we we do like the metaphor that I like to use is like an agent. And so if we're thinking about a human person, And they're reading manga and then they're like, this is great. I love manga. I'm going to go make some manga. They only do that after reading it. And then they interpolate all of the things that they've seen before in their new creation. So like if you, if that is a right for like a human creator, I think it should also be a right for a machine creator to just be able to read source material and then be inspired by it.

Kerp:
Yeah, that's, that's honestly the most common, um, thing I find myself saying when I'm involved in conversations about this stuff, like different business contexts. Um, Shit, I lost my train of thought. What did you just say? Ha ha ha.

gptboss:
I said that if humans have the right to read media, or like view media and then be

Kerp:
Oh

gptboss:
inspired

Kerp:
yeah.

gptboss:
by it, so should

Kerp:
It's to

gptboss:
machines.

Kerp:
say like, look, this, this forces you to look at the fact that that's not dissimilar from what it is to learn to graphic design as a professional, right? You go to school, you read books. I have some in the other room from my experience in art school where you just go, Hey, this is a type of thing. And that's a type of thing. And then it's all connected. That's what the textbook is talking about. And then that involved to this and that evolved to that. And graphic design is one of my favorite cause includes like. Lithography didn't exist at one point and then it did. And there was a explosion of cool stuff put on posters, you know? Uh, anyway, um, but this is an example of a battleground, right? Where it's like the race condition is not as scary as I need to talk to you about paperclip machines that turn us all into goo, like, um, But there are legitimate things here where, you know, economic leaders are going to go, yeah, we're just going to, we're going to, and, and, and there's a legit reason on the level of it just, it gets weird. Cause it's like, okay, then is that the agent has rights? Uh, can it sign a contract?

gptboss:
I think it's that I have rights as the operator.

Kerp:
Right. And that makes it more like building a website or like a social network in

gptboss:
Totally.

Kerp:
that sense. Right. Hosted content is a part of it. You maybe don't know. Okay. Moving on. Open AI has plans! This one from Hoomin- Hoomin Loop? Hoomin Loop? Hoomin Loop. Open AI has plans according to Sam Altman. Um... Sam came and talked to them and kind of laid out some of the open eyes stuff. Uh, I left the one bullet point in here because I thought this one was, was interesting open AI is heavily GPU limited at present. Um, even at the size that they are, they don't have the access to compute that like Microsoft and Google do to solve scalar problems in the development of like their next model or whatever. That's a real concrete problem, no?

gptboss:
Yes, the unavailability of GPUs. I'm looking through the article. There's some specifics that I would like to talk about.

Kerp:
Yeah, yeah, please

gptboss:
I think that we have time.

Kerp:
call out anything else.

gptboss:
Yeah, so he is, like, they're basically announcing what they're working on, right? So it says right here, open AI is heavily GPU. Yeah, exactly what you're talking about. So in what respect are they GPU limited? Are there too many people querying GPT-4? No. The things that they want to do with this extra GPU availability, I desperately want them to do. which is more attention. I talk about when I'm like designing things or developing things for people, I talk about how limited attention is, I still don't have my 32k token context. And like, I'm probably like one of their top customers, right? So this is it's like, there's just isn't enough GPU for like longer attention. What longer attention allows you to do is like do document QA over books, right? You could take like a recipe that you were looking for was like, let me take 10 years of tweets from my CEO and then turn that into like a tweet writer that in the style of this person. And there's not enough attention available right now. Our maximum is 8,000 tokens. So he really wants to improve that to 32K, but they have in research up to 1 million. But it depends on having access to more GPU compute. Fine tuning, like a lot of people are asking for fine tuning. This would be another really great example of what you're looking for, taking those tweets and creating a fine tuned GPT-4. isn't available, has never been available, and it's because of

Kerp:
Right.

gptboss:
the GPU bottleneck. And then also cost. GPT-4 is expensive, right? I had a client come up, she does like online tarot card readings,

Kerp:
Mm-hmm.

gptboss:
and then wanted like an interpreter of the tarot cards for the GPT-4 to say, this is what these cards mean in combination.

Kerp:
Mm-hmm.

gptboss:
And when I told her the price, she was kind of like, oh, dang, I have to severely rate limit how often people can use my app, right? They get like one a day. So there's a lot of use cases that we want

Kerp:
Which

gptboss:
for

Kerp:
is

gptboss:
this,

Kerp:
fair,

gptboss:
especially

Kerp:
that's

gptboss:
for

Kerp:
how

gptboss:
like,

Kerp:
a horoscope works.

gptboss:
yeah, but Taro's different. You

Kerp:
Yeah,

gptboss:
can do

Kerp:
no.

gptboss:
Taro as long as you want with different questions.

Kerp:
Yeah.

gptboss:
So there's like art and leisure activity, right? We're choked on GPU, which chokes what these machines can actually do for us. It has to right now, because it's so expensive, produce an ROI, so we're not able to use it to improve art and leisure throughout humanity. So that's kind of the roadmap, cheaper, faster GPT-4, longer context windows, fine tuning API in that order. And then he's also looking at rolling out a stateful API so that you don't, I don't have to do attention management anymore. Any thoughts on any of that?

Kerp:
Um, they all strike me as necessary, but the GPU one is interesting because that gets them off the hook to some extent, or, or it establishes where they stand relative to competition on the, at the level of improve the model so that the core thing everyone is paying to access is better, uh, that's rough. And maybe this comes down to the, there's no moat situation. just, they are not as big as Google or Microsoft. And so like the partnerships start to make sense when you look at that part of it, I think, um, which is interesting cause it's really concrete. Like this is literally an, it's a problem of electricity and silicone wafers to run it through. Um,

gptboss:
There's just two more

Kerp:
It's

gptboss:
things.

Kerp:
also a good place to go buy picks and shovels

gptboss:
Yeah, there's two more

Kerp:
hashtag,

gptboss:
things I wanna

Kerp:
not

gptboss:
highlight

Kerp:
investment

gptboss:
here.

Kerp:
advice, but everything else is him just like, why combinatoring this, this situation, right? Make it better for people to use it, use it, doing having a stateful API for the thing. It's just a matter of deploying infrastructure that we understand pretty well from a DevOps sense

gptboss:
Yeah. Yeah. Having a database

Kerp:
few,

gptboss:
tied

Kerp:
few

gptboss:
to

Kerp:
months.

gptboss:
a user key.

Kerp:
Yeah.

gptboss:
So there's two more points to talk about here that are kind of exciting. He said plugins, and this is why combinatoring, like you mentioned, plugins don't have product market fit. So everyone was kind of like pogged under their gourd over like, Oh, I could use chat GPT and Wolfram alpha together. And what Sam is noticing is that the uptake of these plugins, and I feel very validated because I was like, nobody's going to use that, right? Like the plugins that were available was like Expedia and Instacart. So you would say, Hey, come up with like a 500 calorie meal for me, and then order it. Doesn't happen. There's only like 100 people out of the millions of users that are doing something like that. So he's gonna roll back on plugins. And what he said was quote, a lot of people thought they wanted their apps to be inside of ChatGPT, but what they really wanted was ChatGPT inside of their apps. And then the final point is OpenAI has stated they will avoid competing with their customers on cases other than ChatGPT. So a lot of people were nervous about like, oh, I don't want to use open AI for any kind of like pipeline because then open AI can just run that. And he's like, we're not interested, right? We just want to do, we, we just want to be the platform.

Kerp:
which still has security considerations, but I think it makes a lot of people feel better in terms of the design space for doing startup things. Some of what they've said there validates what we're up to at Mission Control in terms of helping people use the tool. But yeah, past that it's, it's crazy watching how fast it plays out and the interest level, like the hype cycle is so much louder than every other one, because now all the modern tools exist like this for us to talk about it. But, um, you know, like I said, they're just start-uping, uh, Hey, here's what we're working on, here's the roadmap. Here's what we're seeing in the data, which is everyone says they want to plug in, but you don't know. I talked to a product manager, YC wants about something and they referred to, uh, they referred to everything in product design, past a point as fan fiction.

gptboss:
Hehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehehe

Kerp:
I think about that often. It's like past the point, you're just going, we, I think the user wants, and it's like, you're either gunning toward yourself

gptboss:
Yeah.

Kerp:
or you're just making stuff up and you

gptboss:
Yeah.

Kerp:
have to put things in market and test it.

gptboss:
Yeah.

Kerp:
People think they want plugins. Some of the power you were users maybe figure it out. But if enough people don't go use the plugins, then, then that aspect of it is, is you can it,

gptboss:
specific.

Kerp:
you put those resources into developing a product feature that people do want.

gptboss:
A specific prospect comes to mind, which is a family-owned jewelry store from Africa. It's like direct-to-consumer jewels from the African mines. So it's

Kerp:
Right.

gptboss:
an African company selling African minerals, which I think is cool, good on them. And they were like, I need you to make me a plugin for ChatGPT so that people could buy my diamonds. And I was like, you and every other diamond store. So once it gets there and you're talking to ChatGPT about this supposed wife that you're going to marry and you're buying your engagement ring. You have to pick from like a million diamond stores. So it doesn't make sense for you to pay me the five grand to build this plugin because nobody's like, it's not going to drive traffic for you. Like I had this supposition that the plugin store would be like a new search engine. But I don't see that playing out.

Kerp:
Yeah,

gptboss:
Unless

Kerp:
I'll

gptboss:
we

Kerp:
say

gptboss:
do like...

Kerp:
where I've tried to incorporate it, it's not really like there are aspects where I think it makes sense if you're from like a. Um. like ops standpoint, like there are places where you might want to plug in to make sure that for example, anyone on your team who's using the business edition of chat GPT can only reference existing legal cases or something like that. But yeah, the idea of like trip booking through that interface. It sort of makes sense. But it sure has always seemed to me like it would play out more like This stuff just becomes some version of ambient intelligence, like layered into everything via a tap of, you know, Kevin Kelly's got a good book from like 2016 or something about this. I forget what it's called. Anyway.

gptboss:
Trying to be the dream of like an everything assistant, which I think people

Kerp:
Yeah.

gptboss:
want. They want like one Dunbar's

Kerp:
Inevitable,

gptboss:
number pulled

Kerp:
I think it's

gptboss:
up.

Kerp:
called.

gptboss:
Um, they want one Dunbar's number, like held up by AI. They want to have like their one guy, Siri, Alexa,

Kerp:
Yeah.

gptboss:
whatever it is. And, um, like to do the thing for everything requires this attention breakthrough because to

Kerp:
Yeah.

gptboss:
do tool formers for every activity, like if you are explaining, here's how to use the Expedia plugin and here's how to use the. hotels.com plugin and here's how to use Instacart and here's how to use the gym membership and here's how to use credit

Kerp:
Yeah.

gptboss:
card. All of that

Kerp:
It's going

gptboss:
stuff

Kerp:
to get

gptboss:
needs

Kerp:
rate

gptboss:
a

Kerp:
limited

gptboss:
really

Kerp:
by your ability to teach your team that chat GPT can answer these questions. Like trot

gptboss:
or to

Kerp:
in

gptboss:
like...

Kerp:
a trustworthy way, which is kind of the whole mission control thing.

gptboss:
To allow the agent to select what API it needs to use, it needs to know all of the APIs that it has access to. So

Kerp:
Right.

gptboss:
if you want an Omni tool, you need this 1 million token attention to access every tool. It is possible, but that's what we're waiting for.

Kerp:
Yeah. Although it does sound like they're saying it's only limited by GPU capacity, so. That's that's that's for real. Okay. Moving on to this ever evolving segment that I'm now calling prompt workshop. So I can put like vibey wood in the background on the stream. Um, yeah, again, taking feedback on the fly from the people that are listening to this, uh, the people that we, that we talk to on socials, wherever, uh, They're like, no, I'm, I'm, I'm, people, there are a lot of people right now still at the level of who do I follow to understand I do these things. And I'm always sort of like, here's a couple of newsletters I follow, but you have to just play with it. Um, and I don't think that's for lack of, of thought leaders to find, to talk about this stuff or even like the void we're trying to fill, but the reality is like, you just kind of have to go play with the thing. Um, But also so many people are still starting in a place of like, there are no stupid questions. So we decided to start breaking it out into some broader things that agents like this are really good at, and then talk through in greater specificity, like what that means in the sense of how stuff is, you know, taking shape, um, over time with this kind of stuff. I mean, we can go all kinds of deep if people are into it, but we're going to start at the top. just sort of a broader, like, okay, what's going on here? So. Generative AIs, LLMs, like chat, GBT, Bard, whatever the other ones end up being named as Amazon, et cetera, roll out their plays. Probably going to be Alexa, I guess. Cortana is done. We talked about that before. Anyway, they're really good at summarizing things. Talk about that for a sec.

gptboss:
Sure. Yeah. So I actually I don't know how they do this really. So like just the general way that transformers work is they are like a they're good at like naive transformations of input to output. I kind of talked about this last time I talked about Facebook's image bind. So it doesn't really matter like what you're tokenizing to put in it'll it'll make kind of a good output if the models train correctly and GPT 3.5 turbo and GPT 4 really good at this kind of thing. Basically, it can take a document, convert it into a tokenized representation. So it's like a list of floating point numbers and then run some process on it. And these neural networks are so finely tuned that the output always makes sense. It always matches, right? So as long as you're able to take some text document, it can understand like this is exactly what's in here, but it can point out, it can discover the key phrases. Part of the neural network is able to find the key phrases. So it can cut out a lot of the fluff. and only give you the key important points at the end. It's capable of this kind of naive subtraction of tokens.

Kerp:
Works really well, which is an interesting thing to... Like the philosophical weeds conversation about it is again, how is this different from humans and whatever, right? The ability to have a few documents to write a document while you take some notes and then synthesize that into an explanation of those things. Like there's a lot of rote work like that that happens if you're a lawyer, for example. Um, It's very mechanical. You take a legal writing class and they're like, use these phrases, do it in this structure. Here's the way that you do it. You don't pass. If you don't follow the structure, most of it is made up of citations from cases. Um, That lends itself to a reconstruction without even really knowing what it's doing. Right. Uh, a legal brief, you just, you can ask it for, and it's pretty good at the structure and making sure that that stuff happens, it takes some tuning to get it to, it can write a brief for you, but no surprise that companies like the ones we've called out on here are raising money, doing this in the legal space.

gptboss:
It is

Kerp:
But.

gptboss:
surprising. I can't even talk to an investor. How's everyone raising money?

Kerp:
Uh, but it also,

gptboss:
Yeah, it's totally capable

Kerp:
uh.

gptboss:
of something like this. A similar product that I'm working on is like taking transcriptions of audio and summarizing

Kerp:
I was going

gptboss:
that

Kerp:
to say,

gptboss:
as

Kerp:
so

gptboss:
a document.

Kerp:
talk about it like in a marketing context, right? Like it would have your marketing manager who's trying to optimize. Yeah. This is what I'm doing on an ongoing basis is how can I optimize what's usually a situation of hiring a bunch of contractors to just write stuff for you?

gptboss:
Well, I am literally in the middle of building this

Kerp:
Um,

gptboss:
for us,

Kerp:
yeah,

gptboss:
right? Like

Kerp:
for right.

gptboss:
we're taking, we're taking. So what we do is we take our transcript of, um, this like Riverside, we get this as an MP4 and then that can go to whisper, which is an open AI product that transcribes audio, then that can be summarized and Hey Presto, you have show notes. So you've turned a half an hour, um, podcast into five paragraph summaries. And then it's, you know, it's like a two minute read all of a sudden, right? And so that spreads the ideas, it spreads the news, and it's a more easily digestible thing that can go out as a blog post or it could go out inside of a newsletter, like, and in case you missed it section, this was today's show in a way that's really easy for people to read from source material that is really difficult to read, which would be like a transcript. It's really good at picking out key details out of really poorly formatted stuff, which transcripts are because nobody talks the way that they write.

Kerp:
This is reviewing the transcript of a deposition and then turning it into something is this. Usually done by an associate at a.

gptboss:
And then where this gets really cool is

Kerp:
Ha ha.

gptboss:
semantic analysis. We could do another pass on the transcript instead of just summarizing it, we could say, what was the mood? Could you vibe check this conversation and let

Kerp:
Yeah.

gptboss:
us know if these people like each other?

Kerp:
Oh, that's goggles.

gptboss:
Yeah.

Kerp:
can't

gptboss:
Hehehehe... Yeah... Hehehehe...

Kerp:
see if I can fit them over my CN glasses. I have to wear it out. You need to get contacts. It's, it's degenerated. Anyway, so tune in tomorrow because we'll be getting deeper in the weeds on some of the, some of the stuff Ramsey you're not Ramsey

gptboss:
No.

Kerp:
some of the stuff Mackenzie Mackenzie Mackenzie that's like, I'm halfway there. Ah, uh, come back tomorrow. for some stuff in the weeds on the same topic, like what it looks like to thread together summaries, workflows, and some of the stuff that Mac was just talking about.

gptboss:
Yeah, I've got a very nifty flow chart for you guys. It's beautiful, you're gonna love it, definitely do not miss.

Kerp:
Otherwise, that's Accelerate Daily for today. Thanks everybody for joining us. Like I said at the top, uh, if you got something out of this, like subscribe, even write a review, uh, those metrics really go a lot, a long way when it comes to reaching more people working on the future of AI and the people that we're trying to get to, to have this conversation. So I don't have to do as much production work because if you ask questions, I'll just, we'll just answer them.

gptboss:
Totally.

Kerp:
Yeah. So if you can make the schedule work, jump in the live stream, get in chat. Uh, we'll be sharing it out over mission control socials. So all of us there. That's. where you find your stuff. Thanks for joining everybody.

gptboss:
Yeah, like and subscribe and see you next time.