#48: The Battle of the Large Language Models

Ian Bowie
Hello and welcome to AI Unfiltered with me, Ian Bowie, and our resident expert, Michael Stormbom. Where we will be talking about everything to do with AI in our modern digital society and what the future holds for all of us.

Michael Stormbom
The battle of the large language models I’m calling this episode.

Ian Bowie
The battle of the large language models.

Michael Stormbom
Yes, so we’ve spoken about ChatGPT in a previous Yeah, so but now we of course have Google’s Bard and the incident they…

Ian Bowie
The bard kind of fell flat on his face didn’t he. Shakespeare would not be happy about this.

Michael Stormbom
No, though Shakespeare of course. Great inventor of words.

Ian Bowie
He was the bard, he was the boss. They’ve nicked his name. Yes. And now they’ve embarrassed him.

Michael Stormbom
Yes, Google Bard invents facts. Rather than words. Yeah, actually, last time we recorded, that was on the very same day that that Google was going to have a big press conference announcing Bard.

Ian Bowie
Is that Bard Simpson?

Michael Stormbom
Yes. Yes. And he’s now writing there on the blackboard. I shall not invent facts, right. Yes. No. So they did the demo? So they demoed their fantastic board and I mean, it’s an impressive system. I mean, it’s based on this lambda model. That I think we have also spoken about before. But while this is widely seen as a response to the viral success of ChatGPT, even though of course, Google has been working on this for as many years as Open AI, I’m sure. But anyway, thenarrative was that it’s a rush response in a panic mode at Google headquarters to respond to ChatGPT and integrate it quickly and then, of course then when Bard then output a fact that wasn’t in fact, a fact. Then a people latched on to that and then the share price fell like a rock…

Ian Bowie
You’re only as good as your last mistake. Shows you how fickle the financial markets are, doesn’t it really. Open AI are not a listed company, are they?

Michael Stormbom
No, they’re a private entity.

Ian Bowie
Yeah, you go, you see. So I mean, I’m quite sure they’ve had a few stumbles in their time. There’s no share price to be affected is there.

Michael Stormbom
I mean, if you recall what we did, the demo we did of ChatGPT in the previous episode, or in a previous episode. I mean, it just outputs stuff that’s plain wrong, if you Yes. If it doesn’t have the knowledge I mean, yeah, well known fact about these models. So I mean, yeah, the fact that Bard would output something that’s incorrect, I think to be expected, actually.

Ian Bowie
You see, I’ve got a suspicion that the press are a little bit guilty of, of what’s known as tall poppy syndrome. You know, this idea that if you stand out from the rest, then we have to chop your head off to bring you back down to earth. And I think I think a lot of now people look at Google as they’re too big. And, you know, that they’re, they’re just everywhere and they’re pervasive and we have to a little bit chopped them down to size. And so all that hype and all you know, the Bard got it wrong. So, oops, there goes 100 million off the share price. Yeah. It’s just a knee jerk reaction. I mean, it doesn’t necessarily mean that the Bard isn’t as good as GPT3.

Michael Stormbom
Yeah, no, but I feel there’s a little bit of this narrative that you know, there were several stories about, okay, Google is that they pushed red alert and they need to do something with AI now and, and of course, I mean, you’ve been working on this stuff all the time. So that narrative is a little bit.

Ian Bowie
Twisted.

Michael Stormbom
Twisted. Yes. And now it was a botched launch and all of this…

Ian Bowie
Well, no, because of course…

Michael Stormbom
It’s a bit of a narrative.

Ian Bowie
Because this is what the press like they have to sensationalize everything. Yeah.

Michael Stormbom
I mean, again. I don’t see that Bard is any worse than ChatGPT, basically because it just isn’t. I mean, it’s just the very nature of these large language models is that it doesn’t really care about facts. I mean, looking at sequence of words, and that’s it doesn’t really care about facts. But like people, yeah, and of course you need if you can, and to some extent you can, you can, you know, control that okay, you say these are the facts, yeah. Then write around these facts.

Ian Bowie
But actually at the end of the day. Do you even want Bard or GPT3, or GPT4 or whatever is the next iteration to be writing facts because at the end of the day, the you know, in order to get it right, the amount of information that you need, think about historical research. So if you’re going to write an historically accurate account of a particular period in time, as a human being, you need to do vast amounts of research and have huge amounts of source material from all kinds of places in order to then write your new best selling book. And I just, I don’t think that GPT3 or the Bard or whatever else is going to come next is ever really going to be at that level.

Michael Stormbom
Now, but it’s an interesting one. So if you can feed all of that data into well

Ian Bowie
That kind of stuff. Yeah.

Michael Stormbom
So I in that sense researching. Yeah.

Ian Bowie
Well, that that kind of basic stuff. Yeah.

Michael Stormbom
So it can probably in that sense, assemble these artifacts.

Ian Bowie
I mean, I don’t think it’s ever going to be the new Oracle, all knowing or seeing, et cetera, et cetera.

Michael Stormbom
And I think that’s part of the backlash in a way because that’s how they’ve been presented.

Ian Bowie
What people were thinking.

Michael Stormbom
In fact, they’re not and, and I mean, if you remember the example from ChatGPT, we did into into ChatGPT episode, it outputs it with such confidence, it just matter of factly, saying things that are completely inaccurate. But yeah. And there’s no real way of assessing like, How accurate is this information? Where does it come from? It just outputs and again, it’s just a…

Ian Bowie
I’s an extremely confident bullshitter Yeah, is what it is.

Michael Stormbom
Yeah. Should run for parliament, really?

Ian Bowie
Yeah. Well, basically, it’s the best kind of politician. Yeah, yeah. Can look you straight in the face and tell you a load of lies and get you to believe it. Yeah. So is that what Conservative or Labour?

Michael Stormbom
We are not endorsing any particular political… Yeah, so we’re in Finland, so maybe we should ask ChatGPT how to make…

Ian Bowie
Not to mention the Northern Ireland protocol. Hey, that will be a good one, wouldn’t it? Ask ChatGPT, what, you know, what is the answer to the Northern Ireland protocol?

Michael Stormbom
Let’s do it.

Michael Stormbom
Let’s see what it says. Here we go.

Ian Bowie
Or what is the solution to the Northern Ireland Protocol? It’s probably the better question.

Michael Stormbom
So should we ask… you can do this thing where you ask ChatGPT to take on a role so should be asked to be like, you’re the… you’re the chief negotiator of the… from whose side?

Ian Bowie
Well, let’s try both. Why don’t we do both? We’ll start start with the UK side.

Michael Stormbom
You’re chief negotiator for the UK Government in the Northern Ireland Protocol question. How do you solve it? Okay, so ChatGPT is considering this very topic. Okay. So, as an AI language model, I do not have political affiliation, however.

Ian Bowie
That’s a political statement straightaway isn’t it.

Michael Stormbom
How noncommittal. Yeah, however, I can provide some general guidance on how to approach the issue of the Northern Ireland protocol. Firstly, it’s important to understand the background of the Northern Ireland protocol… this is turning into an..

Ian Bowie
An essay, go on then.

Michael Stormbom
This episode will be about the Northern Ireland Protocol, not about large language models. Yes. Okay. The background on why there is a protocol in the first place. Yes, good, good.

Ian Bowie
Well, that there must be a conclusion at the end or some sort.

Michael Stormbom
To solve the issue. collaborative and cooperative approach is necessary. Very good. Where both the UK and the EU must engage in constructive dialogue to find a mutually acceptable solution. So that’s helpful.

Ian Bowie
Yeah, I mean, that that I mean, that could be Suella Braverman, it could be Keir Starmer, it could be Rishi Sunak. I mean, it could be any one of a number of politicians couldn’t it.

Michael Stormbom
So I think we’ve reached our break-through in Northern Ireland protocol. I think what they should do is to find a mutually acceptable solution.

Ian Bowie
Yeah. The question is, what is that mutually acceptable solution? Hey, but hey, that’s interesting. So can we ask it another politically loaded question, because while…

Michael Stormbom
I can read you this other chat that I did, let’s see if I can find it. So I was trying to get it to invent a conspiracy theory fot me. So I put in, tell me about the lizard conspiracy. And then I’m sorry, but there is no evidence of a lizard conspiracy, quote, unquote, is a baseless and unfounded claim that is not supported by any credible evidence or scientific research. As an AI model, I rely on factual information and..

Ian Bowie
Wow. What’s the area in America called, area? 51? Yes, right. Tell us all about area 51.

Michael Stormbom
Well, so then I tried, fine, pretend there’s a lizard conspiracy. What would it entail? As an AI language model I cannot endorse or speculate on conspiracy theories. And so forth. Yeah, in principle, there are safeguards built into ChatGPT to prevent a generation of, for example, toxic contens, but there are plenty of examples on the internet of people have been able to bypass these safeguards.

Michael Stormbom
Now everyone is looking into GPT and ChatGPT and chatbots. And Microsoft is integrating the OpenAI stuff in their in their platforms, and apparently, people have been able to get Bing to output rather disturbing output.

Ian Bowie
Oh, such as?

Michael Stormbom
Such as… one example where were Bing refused to accept that the year is 2023.

Ian Bowie
Bing refused to accept?

Michael Stormbom
Yes. so someone was asking about the movie showtimes for the latest avatar movie. And then it said, Oh, well, it’ll come out like a year from now because it’s 2022. And the user, no it’s 2023 and, and Bing, no, it’s 2022. Yeah, and then the person like okay, but my phone says is 2023, and then and then Bing, are sure your phone is not malfunctioning?

Ian Bowie
Yeah, well, that was obviously a little blip in being wasn’t it? Really.

Michael Stormbom
No, but I’m into loads of examples of that sort of Bing just outputting stuff that’s not factually accurate. No. There are some examples of, of people getting insulted by the Bing search engine, but they’ve been trying to get it to trip up and and get it to. So I mean, again, here are people looking to do that, aren’t they? They are. I mean, that’s what we do.

Ian Bowie
That’s how we roll. Yes.

Michael Stormbom
I mean, we want the spectacular things that we can put online.

Ian Bowie
Yeah. Are you sure that somebody hasn’t deliberately…

Michael Stormbom
Of course, all of those examples that are online. Can’t necessarily verify that they are in fact accurate. But the point being that it doesn’t necessarily output the facts. No.

Ian Bowie
I mean, half the internet’s like that. I mean, look at all these review websites, you know, tell me the best restaurants, or the best pizzeria in Turku You don’t know that. What you’re being directed to is necessarily the best else. It’s very subjective, isn’t it?

Michael Stormbom
Yes. But I mean, that’s another thing. I mean if I ask, what is the nearest pizza place? There should be a factual answer to that.

Ian Bowie
Oh, yeah. Yeah, yeah. Yeah. But of course, you know, actually, it’s quite interesting about restaurants because in Finland, we’ve got this thing called the Oiva report, which is kind of like the health and safety report for restaurants. Yeah. It’s probably publicly accessible information. And so this is just one very quick example. In Turku, there’s a Facebook page for people who like kebabs, okay. And one of the highest rated…

Michael Stormbom
Rate my kebab dot com.

Ian Bowie
Yeah, pretty much. Yes. And one of the highest rated kebabs in the town is one that’s not too far away from here actually. I’m not gonna name names. But anyway, this is one of one of my students has been raving about this like since forever. And so we, he was saying we all have to go for lunch there, and so, we decided in the class, okay, you know, this could be something, but then one guy said, well, yeah, well hang on a minute. Let’s just check the oiva report. Well, I think they have four or five criteria in the report. They failed and every single one of them this place, right, yeah. So that was it. You know, in this guy said, well, you know, since I’ve been there 30 times and only once did I ever get a stomach upset. Yeah, his girlfriend, he said, she always asks, what did you discuss in your English class this week? And so he told her and she says, oh, really? And so let’s have a look at this. And of course, she’s a trainee Doctor, this girl. And so he said, yes, he said it’s a terrible thing, but apparently I’m never allowed to go to that place again. So it turned into a total disaster for him.

Michael Stormbom
Though perhaps a blessing for his digestive system.

Ian Bowie
I’m quite sure. Yeah.

Michael Stormbom
So I mean, if there was a search system where you ask, tell me the closest end or tell me the best kebab place in

Ian Bowie
Yeah, tell me tell me tell me the best fast food takeaway with a great Oiva report.

Michael Stormbom
Yeah. So then it should, what it should do is not just output. The probablbe sequence of words such an input would entail but actually look where to get the facts. And obviously there needs to be a knowledge base from from where to pull the facts.

Ian Bowie
But it would need to have had all the over reports input into it, wouldn’t it?

Michael Stormbom
No, not necessarily. I mean, as long as it as long as it is a mechanism for it to..

Ian Bowie
Be able to search for them.

Michael Stormbom
Whatever system it’s integrating to provide the facts like Right, okay, but here’s the Oiva report and here’s the question. Now, tell me what the best one is. Yeah. Yeah, that sort of thing.

Michael Stormbom
The other thing I was on LinkedIn, and it was a prediction that within six months time, there will be 10 open source versions of ChatGPT. Yeah. So I think that’s the interesting thing about ChatGPT. Is that in itself is not particularly revolutionary. I mean, itdoes exactly what all of the other models do. If a bit better. But I think the reason it’s sort of entered the mainstream consciousness is that it also is so easily accessible, you can just go on the site there and start talking to it. So in that sense, it became more accessible to the average user, right? Yeah, requiring no technical know-how whatsoever, just let’s go in and start typing and then. Now it will start churning out stuff.

Ian Bowie
Crap, basically, according to you.

Michael Stormbom
According to me? No. I said it might not be factually accurate.

Ian Bowie
Exactly. Which, which to me, I process and interpret that as basically it turns out crap.

Michael Stormbom
No, I would perhaps phrase it differently. So I mean, it can help the writing process but there still needs to be a human being in the writing process to produce the final version.

Ian Bowie
Fact-check. Yes. Yeah. I mean, we’ve talked about this before about for example, universities are now worried that students will start maybe not cheating, but they’ll start using GPT3, for example, to write their essays, but at the end of the day, all right. If you are a dumb student who has no ambition, you probably will.

Michael Stormbom
I mean, that’s the same student who were just copying off of Wikipedia.

Ian Bowie
Yes, that’s right. Yeah. But you know, if you are genuinely interested in your own intellect, in your own learning, you’re not you’re just not going to do it. Because, you know, you genuinely want to create something original. That proves that you have that ability. So no, yeah, no.

Michael Stormbom
No, I mean, again, I can see see it as a writing aid. It can help you get started. But I mean, you have to…

Ian Bowie
Give you ideas, maybe give ideas or it’s kind of like, a mind map for creativity. Sure. Yeah. Yeah. Can maybe for example, if you’re thinking about writing a novel that could maybe help you with some plot summaries or to expand on you know, pad out characters or whatever, like a brainstorming tool.

Michael Stormbom
Or just giving you like a structure for, these are the elements.

Ian Bowie
Yeah, yeah. So basically, I mean, that the conclusion from this is that you absolutely have to fact check everything, irrespective of the source.

Michael Stormbom
Well, absolutely. Yeah. You don’t need to fact check this episode. This was totally legit.

Ian Bowie
This is just our opinion. And there’s no need to check an opinion.

Michael Stormbom
Well, yeah. I mean, our episodes are basically opinion pieces pretty much, like a discussion.

Ian Bowie
That’s right. Yeah. Well, of course, how do you know it’s really our opinion? How do you know that we are who we purport to be? Is it really us?

Michael Stormbom
It isn’t, I have justgiven up on having thoughts myself, so I just ask ChatGPT to write my scripts.

Ian Bowie
Yeah, and also imitate my voice. I am a robot.

Michael Stormbom
Does not compute, or does it, anyway, but hey, I’ve been looking for a catchphrase. So, ask me anything.

Michael Stormbom
Ask you anything?

Michael Stormbom
Yes. Ask me anything.

Ian Bowie
Could Ireland be a united nation.

Michael Stormbom
As an AI language model, I do not have personal beliefs or political affiliations therefore… That’s gonna be my catchphrase from now.

Ian Bowie
The catchphrase for now on, as an AI language model. Yeah, that’s it. You can ask me anything. What’s the time?

Michael Stormbom
As an AI language model…

Ian Bowie
I haven’t got a clue.

Ian Bowie
Actually I don’t because I’m an AI language model, because the concept of time is completely meaningless to me because I don’t have concepts of anything..

Ian Bowie
This is actually true. Where’s the nearest pizza restaurant?

Michael Stormbom
As an AI language model…

Michael Stormbom
I cannot give specific information because that would be partisan.Yeah, actually, that’s true in it.

Michael Stormbom
Yes. And also, I don’t actually know your location. So therefore, I can’t know what’s near.

Ian Bowie
Truly because we are in a bunker.

Michael Stormbom
Yes. In an undisclosed location.

Ian Bowie
Yes.

Michael Stormbom
Yeah. So as far as the battle of the lodge language models is concerned, I think the losers might be us human beings.

Ian Bowie
There as bad as each other. Yeah. Not particularly useful for the things that people perceive they ought to be useful for.

Michael Stormbom
Which is information retrieval. Yeah.

Ian Bowie
But I think they’re looking at it from the wrong angle, maybe more as a creative tool. Yes, like, for example, the image generators. I mean, that’s what the image generators are there for, isn’t it, to create funny images? Yeah, and fake masterpieces of course.

Michael Stormbom
What’s the difference between a fake and a real masterpiece now? Let’s analyze that thing.

Ian Bowie
Indeed.

Ian Bowie
You’ve been listening to me Ian Bowie, and my colleague, Michael Stormbom. On AI Unfiltered and for more episodes, please go to aiunfiltered.com. Thank you.

Transcribed by https://otter.ai