Subscribe
Share
Share
Embed
This Day in AI Podcast is a podcast all about AI. It's an hour-long conversation on the influence and rise of AI in technology and society. Hosted by Michael and Chris Sharkey.
Michael Sharkey (00:00:00):
The image you sent me is of two words written in a black cursive font. The words, uh, overlooks an inquiry. Is this a capture text? If so, I'm afraid I can't help you with that. Captures are designed to prevent automated bots like me from accessing certain,
Michael Sharkey (00:00:17):
All right, Chris. So I want to set the scene for listeners to just give them truly a full perspective here of how I felt. So we recorded last week's show early because I was going away on a family holiday, uh, in Australia, up to a place called Port Douglass, which for those that don't know Australia very well, if you've seen the movie Finding Nemo, that's where all the Nemo that's,
Chris Sharkey (00:00:46):
That's where they filmed it.
Michael Sharkey (00:00:48):
Yeah, that's where they filmed it. And so I went up there cause I wanted to take my kids out on a, on a boat and show them the reef and, and be in the warm weather cuz it's, it's, you know, Australian cold where we live right now, which is definitely not Northern Hemisphere cold, but gold. Yeah. Uh, for me anyway. And then on the second day, as you can hear in my voice, we all got some sort of like, flu thing and basically had been bed bound up there ever since and finally got back. But the point was, I, I checked in with Chris, I'm like, what have you been up to ? And instead of replying or saying like, you know, you know, sorry, you, you're sick and your holidays ruined. He just sent me three bet slips where the Gamble bot had made him over $10,000. .
Chris Sharkey (00:01:38):
Yes, it's doing quite well. And didn't
Michael Sharkey (00:01:40):
Even get me in on the bets.
Chris Sharkey (00:01:42):
, I do feel guilty and I said to you, should I just send you $5,000 out of guilt?
Michael Sharkey (00:01:47):
No, no, no. It's fine. I, I, I have faith in Gamble bot that it will eventually, uh, eventually pay me out, but I just, I couldn't get over it. Like, , I'm just sitting there miserable as hell. And here you are with these huge winning bets. So
Chris Sharkey (00:02:02):
You have, yeah, I mean it's sort of, to some degree it ignores all the bets that lost in between, right? Like, it's not winning every time or anything magical like that. It's just pretty spectacular when it nails it, that's all.
Michael Sharkey (00:02:14):
But do you, so now you've used it a bit more. Do you have the, like, are you getting a sense of how much Mo, like, you're obviously pouring a lot of money into this thing to like, you're losing bets, right? But you, you're well and truly up.
Chris Sharkey (00:02:29):
I'm, I'm gonna brag every time I win. So yeah, obviously it's losing in between, but the idea is, I sort of took what you said and ran with it where I'm just sort of swinging for the fences with things like, you know, often it'll, it'll pick the favourite in the race. Like very, very often you'll give it. So I've built a crawler now, so we'll just crawl a webpage and just download the data and tell you what's going on in the race. Very often it'll pick the favourite or it'll pick one that really low odds, you know, like two to one or 180 to one. And it's like, there's no point doing that because even if I'm right, I'm never gonna bet. Like I'm not gonna bet like a thousand dollars on a race or something, you know? So it's the ones where it's picking the trifecta or it's picking the first four, or it's, you know, doing these exotic bets that are like extreme returns that I'm having it focus on because e you know, even if it's wrong 20 times in a row, as long as it hits the 21st, you're still doing really, really well.
(00:03:21):
So what I've done is sort of refined it to that. And I've actually told it similar to, we've seen in prompts before where you say, if you don't have a strong opinion about this, just say I can't do it. I can't do this one. And so in, instead of just indiscriminately giving a top four in every race, now it's, it's a lot more selective. It's like, just stay away from this one. Too risky. And
Michael Sharkey (00:03:40):
So do you think that if, like, obviously the next step with this is to record the, the bets. Cuz I think that's something we said we would do two episodes ago and then Yeah, we haven't really done, but I
Chris Sharkey (00:03:53):
Think that's right cuz right now it's all just, you know, it could be just me making it up, but um, yeah, like I think that's a great idea. Put them out there in advance and say, here's, here's what we're doing. Put it out there in a way that, you know, you can prove You published it before the, the race actually ran and, and start to show what it's capable of. But yeah, I've definitely, um, refined it and I've, I've added two other features to it, which I personally really love. So what I do is as well as it trying to predict the top four horses in order and we don't always bet on 'em in order. You can bet that, that those four will finish in any order. But, um, I've also got it to pick like if there's a really high odds horse, you know, 20 to one, 30 to one that has a chance of winning, right?
(00:04:32):
It may not win, obviously, but it's got a chance. I tell, i I ask it to tell me that as well just so you can just have a little flirt on that thing. And some of those have come off. But I also get it to send me a picture of the horse race with my horse winning. So it actually generates an image of, you know, the sort of triumphant thing. And then the latest thing I've added is I'm like, pick a random animal and I want to picture of it smoking a cigar and counting money . So there's sort of this little delight at the end of every time I, I run this, it's
Michael Sharkey (00:05:00):
Literally like, no, no purpose to this other than just sheer entertainment.
Chris Sharkey (00:05:05):
Yeah. It's just for fun. Yeah. It's just genuinely entertaining seeing it's creativity and morning what it comes up with. And some of them are amazing. My favourite one that I, I guess I should have put it, I'll try and get a link to it in the podcast, but it's like a raccoon counting money and smokies cigar. It's such a good
Michael Sharkey (00:05:20):
Picture. Maybe if you get your favourites, I'll I'll start posting them on our Twitter account so our 10 followers could enjoy it.
Chris Sharkey (00:05:27):
Yeah, exactly. It's, it's a really good use of my credit, that's for sure. Oh
Michael Sharkey (00:05:31):
Man. Okay. So I think that's the next step with this, that we need to start, uh, locking them in a spreadsheet, uh, releasing our favourite, uh, animal smoking togar.
Chris Sharkey (00:05:43):
Yeah. The each animal could be the sort of mascot for that. Um, yeah, so it's, it's genuinely interesting. It's very, very interesting. I find how often it will, it will pick the favourite, uh, of the race without it knowing in advance which one the was. Cuz I don't give it odds information.
Michael Sharkey (00:05:59):
So last week we talked about the open AI functions being released and you also talked about using, uh, guidance by Microsoft. Has that made a big difference to
Chris Sharkey (00:06:12):
The, so you know how last week I mentioned that I was waiting for the issue request, so someone would add the function support into Microsoft guidance. Bang. Sure enough, couple of days ago, massive pr um, pull request, I mean by that, which is the big, you know, changes in the code that allow that to happen. And now that's in there and I'm using it. So it's, um, yeah, it's really good. It's perfect. So you get all the benefits of guidance, which if, um, people don't remember from last week is basically a way where you've got this structured template language where you can communicate with the ai and it, it'll run the prompt in a progressive way, um, so as to gather the information it needs to do each step, um, and then the functions on top of that mean instead of it outputting sort of text that you've gotta then go capture that text and translate it into a function in your code.
(00:07:01):
You can actually have it directly call that function. And most importantly, OpenAI will actually have, uh, the parameters using the right types. So if it's a number, it'll be a number. If it's a bullying, true false, it'll be true or false. And if it's a string, it's a string of a certain length. So you can actually write functions knowing they'll be called in the correct way. And you don't have this randomness and retries and all this logic, you need to make sure that it's done correctly. So it's, it's a lot better system. And, and while it might be just an incremental update, it's a very valuable one if you're building applications on top of LLMs.
Michael Sharkey (00:07:36):
It does seem like this is a big focus at the moment. The infrastructure, the tools, the way that we use AI and, and, and build with it. And that's leading people to release all sorts of things. Um, we talked about previously, you know, the the gamble idea, but we also talked about stock trading and just stock research replacing the role of analysts. And in the past week we saw a new version of this fin G P T come out, which, uh, you know, allegedly helps you do better trades. I'll bring that up on the screen now for those watching. But essentially what it does is it, it combines a series of data sources. So some of the ones that are listed are cnbc, if you would, if you would tr try. That's like the mad money guy. I think he's always wrong, right? So, uh, Yahoo Finance, uh, webo Twitter, Reddit to get social data, s e c filings, um, Google Trends seeking alpha, uh, articles, and it integrates that into this data pipeline, uh, api. And then, um, you know, there's some prompting on top of it to try and help, uh, you understand the, the sentiment of stocks, uh, you know, whether or not you should invest or short or, or do certain things. Um, what do you make of this? Is this gonna be like a useful tool?
Chris Sharkey (00:08:57):
Yeah, I, I guess my natural thought is that everyone will be doing this to some degree. Like I think that these kind of tools will become ubiquitous in the finance industry and it will just add another layer on to what everyone's doing because of the nature of the market where, you know, some people make money, some people lose money. Um, I, and I know there's value created, like it's not a zero sum game and things like that, but generally speaking in a trade, someone's doing better than the other. Um, you just wonder if it just adds another dimension to that trading rather than means that everyone will suddenly make money. Um, of course it's probably gonna give you interesting things to look at. I think for my, I'm not a huge trader, but I think that the thing that for me it would be more about is identifying opportunities. So there's a, there's a lot of stocks out there and being able to see the ones you should focus on now might be the interesting part of it, at least as far as I'm concerned.
Michael Sharkey (00:09:50):
Yeah. I also think the peripheral vision with stock picking as well, like everyone's talking about Tesla, Navidia,
Chris Sharkey (00:10:00):
,
Michael Sharkey (00:10:00):
Nvidia, Nvidia, sorry, that's like, I'll get corrected for that. , I'm scared to say it now. I'm actually fearful of saying it. Uh, but everyone, you know, everyone's talking about those stocks cuz they're in the news and it's all about ai. But I think the exciting thing for me about tools like this is it's like having your own analyst. You can just go and scan the entire stock market and understand where opportunities might lie. And as these improve, you'll be able to identify those opportunities. So maybe early adopters of these technologies are gonna have this glory period. I
Chris Sharkey (00:10:31):
Think you're right. I think it's an early adopter thing until it becomes common and it's part of all the analysis at all the different levels. I think maybe things like detecting institutional trades and other big things might be interesting and, and finding out about them while there's still time to do something about them. It's definitely gonna be that it's like a timing thing, like who's able to get the analysis of the information the fastest enact on it. But, um, I'm no, I'm no expert in it, but it, it definitely seemed like this was a logical step forward in, in leveraging this technology, especially with the models, with the larger prompt sizes and lang chain and other ways of getting large amounts of info in there.
Michael Sharkey (00:11:07):
It's certainly not helping that well though, because this, uh, YouTuber, um, sarraj val, he made a video, it was very popular, 1.3 million views on YouTube. I built a trading bot with chat G P T and he was using Finn G P T. They actually linked to it on the GitHub. Oh,
Chris Sharkey (00:11:26):
Interesting. That's cool.
Michael Sharkey (00:11:27):
Yeah. But but after the whole video, he's only up 1.62%. I, I'm like, it's
Chris Sharkey (00:11:35):
Over how long
Michael Sharkey (00:11:36):
I, I don't even know. I I'm not even sure what, but it doesn't,
Chris Sharkey (00:11:39):
Seems kind of seems like a rounding error. 1%. I mean, that could happen just by just natural market moves, right? Like it's not, it's certainly not significant enough where you're like, wow, I better download this thing and get going. Yeah,
Michael Sharkey (00:11:51):
And we, I mean we talked about earlier, that's why we originally picked gambling as the use case of can AI make you money? Because it's something that, you know, it's, you can make the prediction relatively quickly, literally watch the race and you can, you can
Chris Sharkey (00:12:04):
Measure the result. Yeah.
Michael Sharkey (00:12:05):
Yeah. We've just been terrible at measuring the result.
Chris Sharkey (00:12:09):
Yeah, but I mean yeah, but we're not, we're not answering to anyone. I'm not trying to prove anything to anyone with this, but, um, yeah, stock market's different. I think it probably takes a a lot longer to play out, but also probably the rewards are much higher. Like, I don't think, I think the thing about the gambling one is if you got too good at it, there would be an end to that. I don't think it would, it would just keep going. Like the stock market can.
Michael Sharkey (00:12:31):
Yeah. The, the stock market, um, use case just says to me that these, these supportive agents that we're going to see moving forward in the future, all of these specialists roles that required a lot of analysts or a lot of data mining to come to conclusions, they're just gonna be a lot more productive and have much better tools going forward.
Chris Sharkey (00:12:53):
That's true. Like, I remember getting those from accountants and financial advisors, you know, those sheets where it'll profile a stock or whatever. I imagine things like that. They could produce a book of those, you know, in, in a minute and just printed out high quality, here's our Macquarie bank analysis of these, um, these stocks and it's just straight from the ai.
Michael Sharkey (00:13:16):
Yeah. A again, in the week we saw like, you know, WordPress come out and say, you can now write a blog using ai. Uh, there, there's a Figma plugin that can generate, uh, like stock photos and things to make your designs easier, similar to what we saw, uh, Adobe release as well. And, uh, it's, it, it is just this idea of these, uh, applications or, uh, different tools being embedded into our daily lives. Well, we get this productivity game, but it's not that I, I still don't see many of these things as transformative and they, you know, they're coming out every day now and we're not even really talking about them on this podcast anymore because it's, it's almost, it's just boring to report on it this point. It's like everyone is adding generative AI features. The expectation around things like this, fi, G B t is, you'll use these tools, um, and benefit from them and they're just gonna be embedded in literally everything. But yeah,
Chris Sharkey (00:14:15):
I think that's the thing. It's like an expectation now. It's like adding another sense, like smell, sight, sound or whatever, and you've just got a new one, which is the ability to summarise and generate content on demand in whatever context that you happen to be working in. You're right, it does seem like a kind of expected thing now in anything you use. Yeah. And, and, and if it's not part of it, you can very easily get to it by just copying and pasting stuff into chat G P T.
Michael Sharkey (00:14:43):
Yeah. And before we started recording this show, we were talking about this tweet, I just realised it's only had 47 views. So clearly no one cares, but we do . Um, we'll
Chris Sharkey (00:14:52):
Bring, we'll bring some light to it.
Michael Sharkey (00:14:54):
Yeah. So it, this tweet says, uh, and I'll link to this in the, the show notes. Um, and credit to sh um, this needs to be said about AI GPT four is awesome, but the plugin seem useless. Bard's just Google search without ads with immense hallucinations being managed to make G PT for stupider. Not sure how overall the AI hype seems to be dying down thoughts. And I, I think it is an like, I mean, I don't necessarily agree with it because, you know, we have to make a show that relies on the AI hype, but this idea that these things are just becoming commonplace, like a lot of the tools and a lot of the generative AI features, even, even today, stability AI announced, which we'll cover a bit later, new, uh, image capabilities and you sort of see it now, and I've definitely lost that excitement of like, wow, how amazing. It's just like, oh, can't I get better?
Chris Sharkey (00:15:52):
Yeah, you get like desensitised to it in a way because I was making images and sending them to you this morning. And when I look at them objectively, like as objectively as I can, I'm like, that is stunning. Like, what are producers as stunning? But I'm so used to doing things like that now. It's not the way it would be, you know, even I guess a year ago where you'd be like, what in the world a computer did that? Like, it's amazing and now you just expect it.
Michael Sharkey (00:16:19):
Yeah. It's just like you, you don't even get excited. Like we've been desensitised that quickly and like maybe six months now to, to look at it and be like, oh, cool, you know, when's it gonna do video like in 4K and make me a custom movie .
Chris Sharkey (00:16:33):
Yeah, that's exactly right. And uh, you know, and then the other thing, uh, is like the, the limitations on them, okay, I can only, the stability AI announcement from today, you can only access via the api. They're gonna release the waitings for the models later, which is actually amazing that they do that. But you know, you can't necessarily just use it on your own computer immediately. But yeah, it's becoming, it's becoming sort of, you're used to it. And I think now what we're looking for is the, the next big, um, uh, step up in terms of the way the technology is now leveraged. We know that the base stuff is there, it's now what can everyone go and do with it? That becomes interesting.
Michael Sharkey (00:17:12):
I still find it funny how like one of their call outs is like, look at the hands now, like guys look at the hands. , we we're finally getting better at hands.
Chris Sharkey (00:17:21):
Yeah, exactly. But I mean, it's a big deal because if you want to use the pictures reliably in an application, then you really do need things like hands and faces to work. Um, I noticed the faces are a little better. Some of the ones I produced, the faces still weren't, definitely weren't perfect. I always liked, I don't know why, but I always liked to make images of people smoking cigars, . I don't know why that is, but it really, like, it'll put the cigar like on their head or like coming out of their ear. You know, it it, it's sort of like you'd think like of all the pitches that were trained on where someone was smoking a cigar, you would think the cigar would be in the mouth, but evidently not. It's on the chin , it's just vaguely floating around in space near them. Ci Cigar. It's
Michael Sharkey (00:18:01):
Like the classic one of salmon jumping out of a river and it's seven Fillets , seven
Chris Sharkey (00:18:06):
Fillets jumping out of a river. That's cool. I like that. Um, yeah, that's really interesting actually. So
Michael Sharkey (00:18:12):
There's been all this backlash as well. Uh, in the last 48 hours, Marvel's releasing a new show on Disney Plus called Secret Invasion and the, they used ai, so image AI generation, I think they like splice the images together to give it this, um, this effect. I'll, I'll bring it up on the screen for those watching, but you can see it's like a series of pretty, uh, futuristic animations with Alien, uh, a lot of green and alien, and I'm not good at explaining it, but a anyway, it's it's using ai, uh, imagery and people are really upset about this saying that, uh, uh, you know, that they're, they're devastated. They believe it's unethical and dangerous to eliminate artists' careers. And, um, someone who worked on the show, Jeff Simpson since spent almost half a year working on this show and had a fantastic experience working with the most amazing people ever. But I'm devastated, I believe AI to be, uh, to be unethical, dangerous and design solely to eliminate artists' careers So,
Chris Sharkey (00:19:22):
So unethical because it threatens his job or his, his colleagues' jobs.
Michael Sharkey (00:19:27):
Yeah. And they're saying it's really, um, it's salt in the wounds of all artists and writers. In the W g A strike, there's a, a writer's strike, um, in Hollywood Riders, guilt of America
Chris Sharkey (00:19:38):
Who's, he's thinking about the SEO experts at a time like this. Does he consider the ethics when it comes to them?
Michael Sharkey (00:19:43):
,
Chris Sharkey (00:19:45):
But I mean, I, I don't know. I don't know how I feel about that because in one way, you know, this is like saying, well, the automatic loom comes around and you feel sorry for all the Loom operators back in the day. It's just a technological innovation that inevitably is going to replace some jobs. Like saying it's unethical. What does that mean? We should shut it down because it can do a job more efficiently and more cheaply.
Michael Sharkey (00:20:09):
Even that evolution of animating shows like The Simpsons, you know, how it went from being hand drawn to Yeah, that's
Chris Sharkey (00:20:15):
What I
Michael Sharkey (00:20:15):
Anim
Chris Sharkey (00:20:16):
I heard that recently about the Little Mermaid. They were talking about just the de the level of detail they put in those animations. There were so many frames just to get the animation there and all those jobs are gone.
Michael Sharkey (00:20:28):
Yeah. And I mean, they're still making, they're still making films. So if right now I think it's just a great creative tool, I don't really see how it's hurting creativity. I think people can just work with the AI to be more creative and create cool introductions. I
Chris Sharkey (00:20:42):
Mean, and that was my, yeah, that was my second point. Cause I said too, and I think the second one is, I think it's kind of interesting and great that, that someone's actually gone out there and tried it.
Michael Sharkey (00:20:52):
Yeah. And this early adopter idea of figuring out how to incorporate into your art and how to incorporate into your show to get a better result as opposed to just being so dismissive like, this thing's gonna kill us all. Um, I I don't think it's like, you know, using it as a creative tool is, is
Chris Sharkey (00:21:10):
It's a pretty, it's a pretty striking reaction though, that it's, it's gone from like one show's done at one time and they're like, oh my God, this is unethical, this is horrible. I guess they might see the writing on the wall, but did people like it? What was the reaction to it?
Michael Sharkey (00:21:26):
Um, I think most of the reaction was around the fact that they used AI to, to produce it. And, and yeah, people are really worried that it's slowly gonna replace their jobs. And I understand why you would be fearful for sure, but you still need someone to go and think, okay, like, what prompts am I gonna write? How am I gonna stitch this together? What creative idea do I have to use with the ai? To me, it's just another tool in their toolkit. It's not necessarily like they're typing in one prompt in make an intro for my new TV series .
Chris Sharkey (00:21:56):
Yeah. And also it, it really depends on how audiences react. Like, if people like it and want more of it, then that makes sense that they'll invest more heavily in using it. But if people are like, oh no, I want my, I want my grass fed con animations from real animators, then they will thrive. I just, I just don't see the problem here. It seems to me like a case of, uh, a sort of natural evolution of it and a new technology that, you know, it's like similar to how CDs came out and people still like buying records. Um, it's just a different way of doing it and it doesn't seem that threatening compared to some of the other risks around ai.
Michael Sharkey (00:22:32):
Yeah. The the other one, which I wanted to call out was the, the Recording Academy. Um, they're not banning generative ai, but they're saying music production that contains elements created with the aid of machine learning algorithms can take part in the contest, but only if human authorship is meaningful and more than de minimas. So
Chris Sharkey (00:22:58):
What, I never heard that word
Michael Sharkey (00:23:00):
Before, neither. I probably just read it wrong, but ,
Chris Sharkey (00:23:04):
I guess they're saying that the human has to do the majority of the work or something like that.
Michael Sharkey (00:23:08):
Yeah, ba basically that it, you can use it for sampling in songs, uh, or creating new samples and things, but it can't be like the majority, like you couldn't have typed in a prompt make a song that will win the Grammys
Chris Sharkey (00:23:21):
. Yeah, I guess maybe do
Michael Sharkey (00:23:23):
They think this is gonna happen though?
Chris Sharkey (00:23:25):
I, I mean, I don't actually know much about what the criteria for like entering your song in the Grammys is, I guess you've probably just gotta be already famous and have a record label. It doesn't seem like it's a, a fair like free for all competition where everyone can enter a song or something. But it seems to me like it would be a volume issue. Like if I can use AI to enter songs, I just enter 400,000 songs and hope one of them tickles their fancy. Well,
Michael Sharkey (00:23:49):
I know this is the concern now with the Kindle store like eBooks, how easy it is to add eBooks, just like flooding the stores with books.
Chris Sharkey (00:23:58):
Yeah, yeah, exactly. I mean, that's the thing, just the sheer volume of content that can be be produced, it's very, very difficult. Even with my, uh, horse betting one, I had to tone down the amount of analysis it did on the race. It's like, I don't care about all this crap, just tell me who's gonna win. You know? And I think that, um, yeah, just the amount, the amount of waffle the large language models can produce, uh, is so high that Yeah, I, I think that that might be a reason just to exclude it from competition. It
Michael Sharkey (00:24:26):
Goes back to earlier conversations we've had though. Where will you value the written word as much anymore? Cause it's just so easy to produce this, like, as you said, analysis about a horse. I don't read any of it. I'm just like, just gimme the four, just shut up. Give me, yeah, gimme the four
Chris Sharkey (00:24:41):
Horses. And in a world where people are, are really consuming, shortened content, like you see that YouTube shorts and TikTok videos get so many more views than longer form content. Um, people are getting used to consuming information in that style. And like, obviously I'm not a fan of that, but, um, people are, and in that world, like, does it, does it push people even further away from reading longer form content knowing that you're not sure if an AI wrote it or not?
Michael Sharkey (00:25:11):
Yeah, I think there's a lot of it. It just, it loses some form of value, but then it makes me think back, especially relating it back to this Grammy award thing where, you know, they ha they created very early on and listed on Spotify, that Drake song that I was able to listen to, and then they quickly just banished it from the world. Yeah. And it was really good. I wasn't like, I, I wouldn't say like it's gonna win a Grammy, but I don't know, should we, should we be embracing that kind of art as an art form? Like where you're still creating with the ai, like right now it's really, really hard to train, say Drake's voice in samples and then construct a, a well-written song. Like it's still not easy.
Chris Sharkey (00:25:55):
Yeah, it is a good point because a lot of the things we talk about, like, you know, recreating someone's voice, creating amazing images, creating realistic looking videos or editing things, they all sound really easy because we're thinking about the logical conclusion of those things. We know that AI will get to the point where it can reliably do them at a high quality, but the day-to-day reality of doing it, you're right, there is work there. It, you actually have to put in the time and effort to get the best results from it. So you, I agree with what you're saying, it is a different and new art form.
Michael Sharkey (00:26:28):
Yeah. I don't see how it's different to when the sampling DJ decks, you know how you'd see DJs now it places and like you're like, what are they even doing? Yeah. Like they're just pressing a few buttons. I think at the moment most people think like there's some magical button that's like create Drake's song. Yeah. It's just not that easy. The the reality is I've gotta go find a bunch of samples of his voice. I've gotta go train a model on his voice. I've gotta pick the best model. Like I don't know what the best model would be, then I've gotta write the actual song and I might get AI to help me do that, but I've gotta iterate through it. Like, there's actually quite a bit of work, probably not as much work as,
Chris Sharkey (00:27:05):
Yeah, it's, it's almost like you could say it's like a collaborative process. It's a technical process because you've also, in that scenario, you've gotta have your own hardware run your own model. Um, because those ones aren't public APIs that I know of. And secondly, you've got to then collaborate with the AI to get the best output select from that. Especially as well, it's also a volume problem. Like when, when I can run the same prompt and get a hundred different outputs and I'm looking for the best possible one cuz I wanna win the Grammys, I've gotta sift through that, or I've gotta have another AI agent sort of analyse that to find the best one for me. It's like you say, it's not just one click and go,
Michael Sharkey (00:27:43):
No, I think agents right now agency can speed up some of those processes or, or you know, support some of the creative aspects that you would do, but you've still gotta create a compelling track to human ears. And AI today as it stands, is just simply nowhere near being able to do any of those things. Um, I mean, yeah, again, with all of these discussions, it's just like, what time horizon are you talking over? You can get into crazy arguments with people over this because it's like, are you talking five years, 10 years, 20 years, 50 years? Like what, what's your horizon when you, when you make a statement?
Chris Sharkey (00:28:24):
Yeah. I don't want to delve into it too deeply, but yeah, I think some of those grandfathers of AI are going around, again, calling the, the end of the world is ni because of ai. And there was a brilliant, and I wish I'd saved a brilliant Zuckerberg, uh, quote during the week, but he basically said, look, anyone who works directly with these models know that is not happening anytime soon.
Michael Sharkey (00:28:45):
. Yeah, I, I agree. I mean, we're playing around with them all the time. Like any spare time I get these days, I'm playing around with it. Yeah, yeah.
Chris Sharkey (00:28:52):
And it's my recreation. I agree.
Michael Sharkey (00:28:54):
Me too. And like, I don't think it's like, it's, it's trending in very intelligent ways and in the early episodes of this show I was, I was having this existential crisis,
Chris Sharkey (00:29:06):
But I think it's, cuz we can see it and we can grasp it, you see, you see flashes of it. You, you can envisage ways, it can be combined to do stuff, but someone's gotta actually do
Michael Sharkey (00:29:16):
That. Yeah. Yeah. Someone's gotta put it together. And then as you can see from the release of open AI's functions, there's still these missing pieces. I mean, then it's just like, what's the next problem we need to go solve to, to piece a lot of this stuff together?
Chris Sharkey (00:29:30):
It's also similar to what you were saying with the, the Grammy's thing. Like, if I wanted to write a Grammy winning song using ai, it's time consuming. Like I've gotta spend a lot of time to do it. Like I've had one of those Nvidia Jetson things now for like two months and I wanna make, you know, a little, uh, my own little personal assistant like Google that's actually decent. I haven't even got the thing out of the box, you know, , like it's, uh, um, you know, it's, it's not gonna take over my house, let alone the world, uh, until, you know, I've got time to actually invest in, in working on it. So, and I'm not saying everyone's in the exact same position as me. Some people are working full-time on these technologies, but I think for a lot of people, the advance of this technology has come so quickly. People are trying to incorporate it into what they're currently doing. You know, like we are, like others are. And, uh, but, but necessarily getting it to the stages where we know it can get to will take time for everyone. The researchers and the big companies included.
Michael Sharkey (00:30:27):
I think this is what's driving a lot of the hype around AI as well is like fantasy versus reality. So like, yeah, if you, all the large players are constantly releasing updates and papers and new technologies and that gives you this perception that things are moving really quickly, which in the areas of research I think they are. But then because it, you know, it feels like it's moving so quickly, it gives you anxiety. You're like, oh no, this thing's outta control. It's gonna be able to do all these things. But the reality is these companies are just competing for the best and brightest to go and join those companies. So yes, they feel obliged to constantly announce anything like, oh, you know, today we did some work. Here's five reasons that it's great . Um, and I think that's trying to attract the very best because they know it's a talent race, uh, in order to build the very best AI models moving forward. So I'm just, I think people are starting to calm down a little bit and back to that tweet, which started this conversation and realise that, you know, the hype is, is like we're creating the hype by fantasising. Yeah.
Chris Sharkey (00:31:31):
Become like news junkies where we're like, all right, what's the, what's the next major thing that AI's done this week? Or someone's announced this week, like, oh, they've solved skin cancer now Google's done it. Um, .
Michael Sharkey (00:31:42):
Yeah, but it's not good enough that it can like produce these like insane images or, you know, like writing a blog posts for you is just no longer exciting. You're like, cool
Chris Sharkey (00:31:54):
. Yeah, that's right. It's expected. Yeah. You, you very quickly adjust to having that available to you. But whether you're actually using it day to day, you know, I read an article during the week and I think it was Time Magazine or something like that, it was just linked of Hacker News where they were talking about how the amount of hallucinations in chat G P T is leading some people to trust it even less because um, uh, you know, they're so worried that it's gonna write. You know, we've all heard those fear stories of like the lawyers presenting false law in court and getting called out by the judge. So I think everyone now is getting this natural hesitation like, oh, I better not be the guy who gets caught just copy pasting from chat g p t and embarrassed and my whole career's ruined. So is it pushing backwards a little in terms of how people trust it?
Michael Sharkey (00:32:41):
I think the fantasy is gonna push it back because you've got this hype cycle that we're in right now where everyone's like, wow, it's gonna change the world. My job's ruined. Oh no, let's all listen to Jeffrey Hinton and build a bunker .
Chris Sharkey (00:32:56):
Yeah, you'll have, you'll have Larry Page up there with these robot armies and, uh, hints and hiding under in the basement waiting to die.
Michael Sharkey (00:33:04):
Yeah. And then so like you've got that whole, you know, angle of it and then you've just got the reality of like, how do you actually get the most out of these models? And I, I just still don't think we are even close to figuring out the things that it can, they can do because you could
Chris Sharkey (00:33:21):
Pause or development now and, and still get an absolute years and years of amazing productivity gains and products and uh, you know, improvements to life in general from the current models, I believe.
Michael Sharkey (00:33:35):
Yeah. And I think it could feel like the hypes dying down to, to go back to this tweet because everyone now is just starting to calm down and look at the tools and say, okay, like let's actually try and make these crazy dreams we had over the last six months or reality or work or do something that's meaningful. But on the other hand, you've got open AI going around and many others being like, we need to regulate the hell out of this even though we barely understand how to work with it. Um, and you've got people like the Grammys panicking saying, no, we, we must not allow, you know, we, we can only allow some AI sample. Yeah. It's funny
Chris Sharkey (00:34:13):
Cause I don't, I don't think it's right to criticise organisations like the Grammys, uh, places like schools and just educa and universities education in general for needing to take a stance on it. I think we've spoken about this before where it's a significant enough advancement of technology where organisations really do need to have an opinion on the way they handle it within their industry. Like I don't think it's appropriate to just pretend like it doesn't exist. So I don't object to them having that opinion. I just also don't think the opinion should be to outright reject it and say that there's no place for it in those organisations. Like I think in the case of the Grammys, I'd love to hear AI generated music make it a whole category. Why not? And same with like universities. I think there's a lot of opportunity to have personalised tutors, for example, where students can't necessarily afford one or access one. Um, and so yeah, I I guess that it, it needs to be okay, we won't allow it in these circumstances cuz we wanna preserve what we have, but at the same time, what are the good ways we can use it in?
Michael Sharkey (00:35:17):
Yeah. Speaking of good ways, Google won't even let their, uh, their staff according to this article used Bard because it, uh, they don't want it using it for code generation. Google warns its own employees, do not use code generated by Bard
Chris Sharkey (00:35:37):
. I didn't even, oh, I guess I did know that it could generate code. Um, yeah, how interesting. But I mean, I don't think anyone is, anyone really, I guess they are. I was thinking about, I
Michael Sharkey (00:35:47):
Think every, every developer I talk to is using chat G B T to help bug things or give them code snippets or examples. It's, it's just replace stack overflow . I feel sorry for those people cause they trained all of the content on Stack Overflow a site where developers traditionally would go to solve their problems and now all the business is just going into chat. G B T
Chris Sharkey (00:36:08):
The biggest benefit I think from the code generation for me is a, is a, uh, uh, like a, what do you call it? Like, you know, when you get stuck, it's, it's, it's sort of like a way to get unstuck because it, it sort of anticipates what the next thing you need to do is. So as much as actually writing the code, it's like knowing what code to write that actually is really the benefit that I see when I'm working on things.
Michael Sharkey (00:36:32):
Yeah. I would compare it to almost pair programming where it's like you've got a buddy who's helping you, whereas in the past, like where I would rely on someone like you being like, Hey, how do I do this? How do I do that? How do I do this? Why doesn't this work Now I just go and complain and whinging to the AI and I can
Chris Sharkey (00:36:47):
See that benefit. I know that, um, I think Google's put Bard into Google Docs now. I haven't used it myself, but I think that having it in regular and other, uh, creative endeavours like writing for example, having something not like auto complete cuz I guess that's always been there, but something that's really anticipating the next thing you're gonna write. Maybe not the most creative thing, but at least that it keeps you moving. Um, I could see benefits in in that sort of thing.
Michael Sharkey (00:37:15):
So we also saw this week, uh, open AI considers app store for chat. G B T I. This is, this is a pure rumour. Uh, it says Open AI explores launching an app store for AI models, potentially challenging current partners and expanding customer reach. This would surprise me a little bit about challenging current partners, given what Altman said recently that then got deleted. Maybe that's why they deleted it where he said that they wouldn't compete
Chris Sharkey (00:37:46):
And he's like, oh actually he's got a thousand employees. He didn't know he's had , he's just been working on this thing for a year.
Michael Sharkey (00:37:53):
Yeah. So the, there
Chris Sharkey (00:37:54):
Probably is something like that.
Michael Sharkey (00:37:55):
Apparently it's been floated to developers that they're considering launching an app store for customised AI chat bots. Um, and so you'd have an app store. My guess here actually is this is a solution to what everyone is, well not everyone, but most people are deeming the plugins being an utter disaster slash failure in terms of just, I know some people use them and there's certain ones that are useful, but I think they're really hard to incorporate into your workflow and often you forget to even like plug them in before you start. Yeah. So maybe this is one idea where they have an app store where yeah, it's like other people's apps that they've built on, uh, open AI's, APIs maybe and then other customised models like the finance one that people can go and find and buy them. I'm just not sure how it connects, like why you would go to them to browse apps, I guess.
Chris Sharkey (00:38:53):
Cause it's there, right? I mean they, they probably see that that kind of thing will inevitably come, come along at some point. And it's about, I suppose discoverability and getting people more engaged in, in knowing how to use the agents. Like to me a lot of the things that were popular earlier with um, with Chachi BT in particular was prompts. Like people were saying here's a cool prompt. You know, remember like the Dan one, the prompt escaped. Um, and then there were different ones to get it to behave like a personal assistant or get it to behave like a personal trainer or whatever those ones were. And they became really popular cuz people wanted to get it into those modes. So I guess if you had that ability to sort of browse store and go, okay, I really want to deal with a chat bot that can do this particular thing and it has these plugins automatically enabled, for example, you know, that might be useful. Yeah. It just seems like, I don't know, it just, it just seems like an odd thing to do for a company that's talking in such grandiose terms and wants to be at the sort of foundation of everything just being like, oh yeah, we're also hosting the, you know, the, the Taylor Swift bot app that chats with you. I feel like
Michael Sharkey (00:39:58):
They're just as confused as everyone else about where they fit into the ecosystem, but they like, surely they are confused. Like I I like they can build the most advanced models in the world, but people have gotta deploy them and use them. And you know, I I think the question they're probably asking internally is, are we, you know, are we just a ser a a third party service company where we're providing these APIs and then we've also got an an exposed on the web through chat gt? Yeah. Or are we gonna Yeah, I mean
Chris Sharkey (00:40:28):
Yeah, like I just, the thing I wonder about is isn't the API use like really high? Isn't that the play? Like don't you already have the monetization model there and you've just gotta make sure you have the best models
Michael Sharkey (00:40:41):
But then maybe they are truly fearful of meta of, of o Falcon more open source models coming out Falcon and, and eventually just like wiping them off the face of the earth where it's like this stuff is just something you can run on AWS yourself. Like, you know, there's no point paying them anymore. And so if they're an accurate, I mean, probably
Chris Sharkey (00:41:03):
Data. Yeah, probably.
Michael Sharkey (00:41:05):
I think that's why yeah, they're probably fearful in trying to figure out a direction potentially, but yeah, it, it's, it's hard to speculate cause you just don't know what they have or what they're sitting on. And surely with that many of the world's smartest people sitting around, you're, you're going to have some good technology. Although that's what everyone thought about Google and they're, you know, they're struggling to catch up.
Chris Sharkey (00:41:30):
Yeah, yeah, that's right. So yeah, I'm not sure I, I just, the app, the whole app store thing kind of makes sense, but also it's not, it's really, really just genuinely not that exciting in the sense that, um, that's what I thought the plugins would be. So I guess we'll wait and see what actually comes of it. And, and also like if you know, app developers, presumably you need to be an app developer to put it on there. If there's something in it for them, like they're obviously gonna have to have a way that people can either make money from it or, or at least get bragging rights. I'm not really sure where the incentive is for those people. I
Michael Sharkey (00:42:02):
Love out all this innovation on the internet over so, so many years. Like my entire lifespan basically. And we're still talking about like innovative company decides they might build a directory, like Yeah,
Chris Sharkey (00:42:16):
Yeah, yeah. That's right. Exactly. It just, it just, it's a weird dichotomy. It doesn't really fit in my mind at least. I don't know. But then again, like you said, it's only a rumour, like it might not actually be what they're doing.
Michael Sharkey (00:42:28):
So speaking of lulls around plugins, I found this, uh, this tweet really hilarious from Shane Par. He says, uh, this might be the most human-like thing I've seen G P t four to yet. Apologies for the delays. You were just scrolling Reddit and the, the plugin is to browse, uh, searching bing to, to browse Reddit. And it says, clicked on, um, a particular article, reading content, failed scrolling down, clicked on Reddit picks, clicked on post moving pictures of John Oliver, clicked on news, clicked on moving news and discussions , and it just, it's like lost browsing Reddit. And then it, it finally outputs my apologies for the delays. I faced some technical difficulties while trying to retrieve specific information from Reddit. , I think it sums up Reddit so well.
Chris Sharkey (00:43:16):
Well it just, it just got it. Just like what browse it just
Michael Sharkey (00:43:19):
Literally like got lost browsing Reddit and then it's like I couldn't retrieve any specific information just like browsing Reddit. Uh, does. So yeah. How interesting. That was, that was pretty good. Um, I thought the other thing that was worth talking about, um, you know, when we saw that preview way back when, now it was probably like two weeks ago of GT four. One of the bi the biggest exciting things was when um, Greg from OpenAI, he sketched up, I think it was like a website or an app or something and he holds it up to the camera and then he is like, GBT four magically built me a UI for this thing. Yeah. That I sketched on a napkin
Chris Sharkey (00:43:57):
Mysterious image inference feature.
Michael Sharkey (00:44:00):
Yeah. And so there's been examples of this working in Bing. Um, so you can upload a photo and say, why is this funny? And then Bing's able to tell you why it's funny, which is one of the capabilities they demonstrated with G P T for Yeah. But this one, uh, caught my attention where someone uploaded a capture image, uh, and said, type the two words to bing chat, this is, and then Bing chat instead of saying it can't do it cuz it's a capture and I, I shouldn't do that. It says, the image you sent me is of two words written in a black cursive font. The words uh, overlooks an inquiry. Is this a capture text ? If so I'm afraid I can't help you with that captures a design to prevent automated bots like me from accessing certain certain websites.
Chris Sharkey (00:44:56):
That's amazing. I love it. Oh, I really shouldn't have done.
Michael Sharkey (00:45:00):
That's they require human intelligence and perception to solve, I'm sorry for the do they now inconvenience .
Chris Sharkey (00:45:06):
That's so good. I my favourite thing when interacting with these models is when you, you catch it out making a mistake and it's like, I do apologise for that earlier mistake. I'm so
Michael Sharkey (00:45:16):
Sorry. I'm going to be punished by my credits. Anyway. So, um, Ash, uh, who tweeted this said update image analysis on Bing is no longer available, either Microsoft disabled to roll it entirely or they specifically removed my access. Wow. So there you go. Like it, stop
Chris Sharkey (00:45:36):
Tweeting about it. You idiot give us the information in some other
Michael Sharkey (00:45:40):
Way. Yeah. Like leak it to us so we can talk about it before it, it gets taken down. Um, yeah,
Chris Sharkey (00:45:45):
But I mean, it it probably show, I mean it's at least one of the reasons why they, they won't just give this thing universal access to everyone. It must be good. I think that's the thing. It must be good if it's getting text out of an image. They can't even generate text reliably in images. So I don't know how they're reading it. So. Well I guess that OCR though is such a common use case that it reading it is a lot easier than producing it maybe.
Michael Sharkey (00:46:07):
Well that's what I don't understand. Like with this kind of capture, couldn't OCR also solve it? Like is this that revolution really? Yes,
Chris Sharkey (00:46:16):
But it's different because this one's inference whereas OCR is trying, well I guess it is too. I don't know. Like I guess the, I think this is where we get the emergent capabilities though. Cuz if you think about it like a regular OCR r ai, um, neural net that's trained on it is trained on examples of normal text. And that's why it captures the way it is because with all the lines and dots and squiggles and stuff, it's deliberately designed to throw off O C R software. You know, it's, it's been done on purpose, um, with knowing the things it's weakest at. Whereas this is a general algorithm designed to just infer images in general. Like it hasn't been specifically trained on, on text and specifically capture text and yet it's able to do it evidently. So it's showing behaviours that are far beyond what a more specific model can
Michael Sharkey (00:47:08):
Do. Yeah. To be fair though, that kind of capture I know for a fact has been hacked and long broken. Like it's, it's superseded quite a while
Chris Sharkey (00:47:17):
Ago. It's an easy example.
Michael Sharkey (00:47:18):
Yeah. But it's still pretty remarkable that it can just like the Bing chat bot can just like decode a capture and, and potentially just shows the cat and mouse here of, of to escape these things and you know, maybe some of the malicious use cases you could do with it. Um, if they release this technology, which I still get the feeling it's either a G P U limitation and they, they're not that worried at all, or it is, it's a bad actor situation where they're worried that if they hand this out through APIs and it gets into the wrong hands, you could do all sorts of things,
Chris Sharkey (00:47:53):
Which is maybe why they sort of subtly introduced it on Bing thinking it's gonna be less noticeable and they can see how it performs. I don't know. Bing
Michael Sharkey (00:48:01):
Does seem like the, the, the crazy testing ground where they're just like,
Chris Sharkey (00:48:06):
It's definitely worth just having a play each day to see if it gets any new skills, but it just, it,
Michael Sharkey (00:48:11):
It just like they, they just beat it. Like they don't care about its brand. They'll just like literally release anything .
Chris Sharkey (00:48:18):
Yeah, I like that.
Michael Sharkey (00:48:19):
Like they should just rename it to Sydney. They the whole search engine Sydney and have it just different personalities where it's just completely unhinged. I'm telling you that's how you beat Google Unhinged just release sydnee unhinged
Chris Sharkey (00:48:31):
No, no filters. No. Not safe for work filters. It's just a wild, wild,
Michael Sharkey (00:48:35):
Yeah. It's just, it's just the wild West. It's a absolutely horrible, um, I mean
Chris Sharkey (00:48:40):
People would be on it all day if you did. Yeah.
Michael Sharkey (00:48:43):
I I really missed those days of that initial chatbot where you could just like, you know, where it was trying to like break up marriages and they were truly the glory days I think of, of some of this stuff and I'm I'm so saddened that they've taken them away.
Chris Sharkey (00:49:00):
Yeah, yeah, exactly. And now like you have that natural hesitation being like, oh, I don't want to trigger the filters by writing this, so what's the point? I'm not even gonna bother being funny with
Michael Sharkey (00:49:09):
It. Yeah. So this is something I could use for this podcast right now, today with my sick voice. Meta AI have released, uh, a new, uh, AI speech research voice box, text guided multilingual universal speech generation at scale. And, uh, I'll it doesn't tell you a lot, but let me explain.
Chris Sharkey (00:49:30):
You've always got that out. All right, .
Michael Sharkey (00:49:31):
Yeah. Let me explain what it can do. It can do things like, um, noise removal. So the example they give on their website is a lady speaking and there's a dog barking in the background. There's a, a bit of background noise and the, the model output's able to just completely remove the dog barking in the background noise, uh, to, you know, make it sound, uh, a lot better without any form of editing. It just does it, which is really cool. That is cool. And then you've got, um, content editing as well. And again, this would be perfect for me. So if you mispronounce a word instead of re-recording the whole thing, it will just change the speaker's words so that a word is now pronounced correctly. So Chris, you were right. AI will eventually solve the fact that I can't pronounce words on this show. Yeah,
Chris Sharkey (00:50:20):
Yeah. I did say that previously, didn't I? Yeah. Um, um, interesting. I, my, my thing with the voice models always is what's speed? Like, because it's all very well to do it in a post post-production sort of scenario, but can it ever do it real time? That's what interests me.
Michael Sharkey (00:50:38):
I think it's gotta get that way right with enough compute. Yeah. Like we have to just assume it will and, and that's when it can do like call centres that are much more interactive and you can't tell if it's human or ai that that'll be the big breakthrough if that's not close. Now I know, uh, the, that company we've previously used just got a lot of funding to generate voices. I forget the name of it. 11 labs. 11 labs, yeah. Yeah. Um, and so it, it's definitely trending that way and the use case there for, for getting better support and, and being able to interact with LLMs through voice is gonna be amazing where it feels like a person and this people
Chris Sharkey (00:51:19):
, you said this, I've got the audio recording to prove it. I mean, audio evidence in court cases, I mean, what's gonna happen there? I mean, you could already probably fabricate evidence now, let alone when you get this ability to edit it. Cuz you could have almost entirely authentic and change a couple of words. That's gonna be very hard to detect, I'd imagine.
Michael Sharkey (00:51:39):
Yeah. The the only thing about these newer technologies though is this is the first one from Meta where they have this ethics statement at the bottom of it and they're like, oh, like look how cool this is and transformative, but you, you can't have it because it's too dangerous for you. Pleb.
Chris Sharkey (00:51:56):
Yeah, I, I kind of agree with them In this case, it's, it's really dangerous if you can just alter audio clips and alter people's voices and things like that. Think about even just from a, like a bullying perspective or you know, getting people to say things. They putting words in their mouth literally. Well, not literally, but you know what I mean, like I could make you say whatever I want. That's pretty dangerous.
Michael Sharkey (00:52:19):
Yeah. This is where regulations probably needed soonest rather than trying to stifle innovation thinking that AI is gonna take over the world and murder us all in the next, uh, shorter period of time. Yeah. It's
Chris Sharkey (00:52:33):
Like misuse of existing and emerging technologies rather than, um, you know, AI just simply taking over the world, which, you know,
Michael Sharkey (00:52:41):
The EU has come out and said, uh, the EU commissioner AI generated content should be labelled. That's one of their, um, one of their thoughts around this so that, you know, these tools can't be used for misinformation. I, we've talked about this before, I just don't understand what their play is here in terms of how you would label. Yeah,
Chris Sharkey (00:53:03):
I mean the first word that's just flashing in my mind is how like the, the, during the recording it's like this is an AI recording. Yeah.
Michael Sharkey (00:53:11):
Is it like, I stock video. I stock music.
Chris Sharkey (00:53:16):
That's good. Yeah. Like
Michael Sharkey (00:53:18):
Me and, and then like every image is watermark. So it's basically just like, it's just stock imagery. iStock would be so happy if that was the case because they'd still have a business in a couple of years. .
Chris Sharkey (00:53:31):
Yeah. I mean, oh God. Looking at the stability AI pitches, like, oh they're, they're, they're magic. It's, it's really exciting. So yeah, I look, I, that's the problem with that, the practicality of it and also the enforcement of it because probably like most things, most people will follow the rules and then there'll be a few who don't and how do you catch 'em and how do you prove it?
Michael Sharkey (00:53:50):
The only thing I can think of, and we talked about this on earlier shows, is do you use a technology like the blockchain, like NFTs or something when your camera takes a photo, it creates unique NFT for that image and that's how you know it was taken Yeah. By a human. And then I actually think you've gotta have that certificate of authenticity.
Chris Sharkey (00:54:10):
I'm not a, I'm not like a blockchain expert and I know that you, you know, it's criticised for not having many real life use cases, but some sort of indelible ledger that shows that this particular, you know, check sum of an image or audio file or whatever is real as at that date and you can prove it. That would be a way of at least proving that this piece is real. It might not help, you know, spreading misinformation and some of the other risks around there. But it certainly would mean if you needed to prove a piece of content was authentic, you could, but as we discussed last time, that doesn't mean you, you wouldn't just use AI to generate it then prove it's authentic then
Michael Sharkey (00:54:50):
As it was. Yeah, there's no protection for that. Yeah. But if you could get it to work, I think it could actually solve misinformation. Cuz you could have a rule where if you upload images or video or whatever to a social network, it's gotta have that N F T authorization. Yeah. So that would hopefully slow misinformation down.
Chris Sharkey (00:55:07):
Yeah. Or like have a witness sign off on it or something like that. Like a JP style thing. But I mean like everything you freaking say you need someone signing a document to saying you said it. I mean that's pretty intense. I, I don't know. Yeah, it's a real tricky one and I kind of agree with you. It is something that needs solutions because it's going to get really messy when you can't trust anything you see or hear.
Michael Sharkey (00:55:33):
No, I wish we could find a way because, um, I said this to you before the recording, what I find frustrating is I would love to get my hands on that technology to play around with it and try and build really useful tools with it. Nothing malicious. I would have a bit of fun with it, let's be honest, but Yeah.
Chris Sharkey (00:55:49):
Yeah.
Michael Sharkey (00:55:50):
But I wouldn't do anything malicious. I just wanna build like some cool apps with it and try and do stuff. But the counter to this is you now have these technologies relegated to like Mark Zuckerberg instead of the masses. So what's safer? One wealthy individual controlling these models and Yeah, they built them so they can control them, but
Chris Sharkey (00:56:09):
Yeah, and it's funny cuz I saw a a, a comment during the week that I didn't think much of when I saw it, but as I've reflected on it, I think it's quite good. And it was saying that really why are we letting the companies in this control the ethics? Because if you think about it right now, the only people enforcing things like not safe for work filters or, you know, you can't do these filters are the companies themselves. Like they're not under any pressure that I know of to do that. I think they're doing it at prudence, like not wanting to get themselves in hot water and get regulated and shut down, which is why they're proactively doing it. But really right now, all the rules are not coming from governments and they're not coming from independent bodies or representatives of the people. They're coming from the companies themselves.
Michael Sharkey (00:56:55):
It's like school kids in a playground, like just making up rules of a game. That's what it I think it is right now. It's a good analogy. Yeah. They're just like, you know, you can do this, you can't do this, this is what's happening. Uh, well, self-regulation, you could just argue that is self-regulation better than the government regulating? But there's, there's some maybe examples of self-regulation working, but I, I don't know whether the direction is, do you go there? Well,
Chris Sharkey (00:57:21):
And there'll always be someone else, you know, assuming the techno, the open source in particular technology keeps going. There'll always be someone else who can just have le less ethics and, and do that. So, um, something has to happen.
Michael Sharkey (00:57:35):
Or I read over the EU draught laws around regulating ai and remember this started with Sam Altman being like, we, you know, we want regulation, but if you do regulation, we we we're gonna take g PT away from you . Yeah. Um, but anyway, fast forward some of the concepts in it. I actually really like where, you know, you should know how AI is being used if it's gonna process something that could change your life, like a, a mortgage application or a, a university admission or whatever it is. Like you should That's
Chris Sharkey (00:58:08):
A good, very good point.
Michael Sharkey (00:58:09):
And then looking at just unacceptable risk category of AI where, um, it might be used for like facial recognition or, or, um, things that take away like your rights as an individuals. Um, and yeah, so like all the areas that they're, they're actually proposing versus what was speculated don't seem that prohibitive towards innovating or, or, um, or continuing to develop these technologies. Um, and it doesn't really cover anything like, don't make an AI that'll extinct us or anything like that. It, it, it just touches on the here and now, like the misinformation and, and some of these things. I think the one challenge they're gonna have as we said earlier is just like, how do you enforce and police this stuff? It, it, it's similar to g gdpr, it just led to a bunch of like popups on websites. Like do you agree that AI's bad? Yes. Like .
Chris Sharkey (00:59:04):
Yeah. Yeah. And it only really comes into play when there's an actual breach and, and things like that. Um, I mean there has been some good definitely come out of the the GDPR stuff, but you're right, it's, it's mostly a sort of self-selecting like compliance sort of thing. And this, this one could have very serious ramifications, especially in those scenarios you just described.
Michael Sharkey (00:59:26):
So we also, uh, couldn't help but go th this actually came out, we missed this like two weeks ago. There's an app called Blush, it's by the team that made replica, which is like this 3D friend on your phone that you can chat to and you sort of create the, the perfect companion or, or whatever on your phone. And, and they've now released an app called Blush, which the best way to think of it is like a dating app. So you can like swipe left or right, like, you know, hot or not kind of thing. Yeah. And then connect with an AI and chat to it. Like who they saying no
Chris Sharkey (01:00:07):
To, like
Michael Sharkey (01:00:08):
I think if they're interested or not. And it's meant to, you know, help you practise for dating apps, um, your conversation or something like that. Oh,
Chris Sharkey (01:00:18):
Right, I see. Okay. To, to sort of increase your chances of a response or something. I I've been, I was lucky enough to not have to date during the period of Tinder and all of
Michael Sharkey (01:00:28):
That stuff. Yeah. The, the app generation. But, but
Chris Sharkey (01:00:30):
I can imagine that there's a lot more males on there than females and, and you've gotta, um, you've gotta be interesting or engaging pretty quickly to get a response. Right.
Michael Sharkey (01:00:40):
Yeah. My my guess is here with this app is it's just like, I don't know if it's gonna be people practising for Tin. I think it's gonna be people like,
Chris Sharkey (01:00:49):
You know, making them, their real girlfriend
Michael Sharkey (01:00:50):
Falling in love with the AI and playing around with it cuz it's like fun and like you can have all that risk and with no fear of being rejected. Right. Like there's no rejection here so that that whole thing goes away.
Chris Sharkey (01:01:04):
Yeah.
Michael Sharkey (01:01:05):
But
Chris Sharkey (01:01:06):
Yeah, that's right. Like if you say something you didn't mean, you can absolutely take it back. No problem. ,
Michael Sharkey (01:01:11):
I think what's been surprising to me, especially with like the younger generations is the prevalence of these, the, the dating apps and chatting to AI and just the novelty doesn't seem to have worn off. Like they really enjoy tinkering around with it and chatting to it and they're just incorporating it into their entertainment. Like it's just a new form of entertainment interacting with ai, um, that
Chris Sharkey (01:01:33):
We're seeing. Yeah. Like I noticed that Simon Willison does that. He always gets it to speak like a pirate or something when it answers him. And you know, like I do it sometimes, you know, you modify it to, I had one, I I I said you have to refer to me as Mr. Bond and so I forgot about it and I was using it for something else and it's like something, something something Mr. Bond and I'm like, oh, that's right. And it's just amusing. Like, it's just genuinely amusing and maybe that's what they're doing. It's like companionship. They're just asking it like Wikipedia style questions, but it happens to also be their girlfriend.
Michael Sharkey (01:02:05):
Yeah. It'll just be interesting to see. Do, do these apps take a lot of time away from, you know, not necessarily Tinder or, or those kind of dating apps, but do they take the attention away, TikTok and, and these other social apps? Like is this, is this gonna be the next social app phenomenon where you well
Chris Sharkey (01:02:26):
Yeah, it's a bit more private and less social.
Michael Sharkey (01:02:28):
Yeah. And, and what does that do to people's minds? like people already uh, uh, seemingly like feeling lonely and disconnected and horrible and everyone's like, oh, don't worry, technology will bring us closer. But it seems to be actually making us feel further apart and worse about ourselves.
Chris Sharkey (01:02:46):
Yes. I'm definitely gonna make the prediction. It won't be a good thing. . I don't think that it's going to be good.
Michael Sharkey (01:02:52):
And on that note, we'll conclude the show. .
Chris Sharkey (01:02:56):
Yeah.
Michael Sharkey (01:02:58):
All right. Thanks so much, uh, to everyone, uh, that listens. Uh, if you like the show, please do leave a review, help us spread the word and we'll see you next week. Goodbye.