Up next in 10
00:00 | 'If you work in advertising, it's over!'
01:16 | Gabriel: 'I used AI as my lawyer'
03:51 | Adam: 'AI taught me algebra'
07:28 | Charles: 'This is why I stopped using AI'
A researcher has resigned from his job at Anthropic with a dire warning that AI companies are "gambling with our lives" as they believe it "could kill us all by the end of the decade".
Jacob Coxon announced his resignation from the AI firm with a lengthy thread on social media, where he claimed that neither Anthropic nor OpenAI were "acting responsibly", as they compete in a race to build bigger and better artificial intelligence systems.
Mr Coxon warned that tech executives and senior researchers will "couch their phrasing" in public to sound sensible, but there is a belief that artificial intelligence "could kill us all by the end of the decade".
Tom Swarbrick is worried, but are LBC's listeners?
Listen to the full show on the all-new LBC App: https://app.af.lbc.co.uk/btnc/thenewlbcapp
#tomswarbrick #anthropic #chatgpt #claude #artificialintelligence #technology #lbc
LBC is the home of live debate around news and current affairs in the UK.
Show More Show Less View Video Transcript
0:00
Yeah, I had no lawyer. I just had AI
0:01
Wow. And I won. I learnt algebra in ten minutes with ChatGPT
0:05
I got on to ChatGPT for two years, and then all of a sudden it lied to me
0:11
If you worked in advertising, and your job was to come up with a storyboard
0:19
for what the advert for this serial was going to look like, whatever, and that's your job, you're done
0:25
you are totally to use the uh the word of the my children and their generation you're cooked um
0:33
because the the clawed are sorted out for you um there are friends of mine who are now building
0:39
really intricate detailed apps for whatever it is um via various websites that allow you to do that
0:47
various uh uh llms large language models that can just build it for you you put in some codes you say
0:52
what you want it to do. You can see that I've never done this myself because I don't know what to do
0:57
So I feel so behind the curve on what is happening. And when you have people who are in roles
1:02
that are quite important in the managing of all of this saying, yeah, we've probably got about a 10% chance
1:07
that we are all wiped out by this. At what point do you sit up and take notice
1:12
And at what point do you ask what is going on? And who do you listen to for the answer? Gabriel's in Hackney
1:17
Gabriel. Hi, Tom. Thanks for taking my call. You're welcome. Metal Gear Solid
1:21
What a great game. One of the best of all time. So I've always said to your researcher before
1:26
I own a small wine bar, and I use AI for literally every daily task
1:32
that I can probably imagine, especially in the admin side of it. You know, I've probably saved
1:35
I would say over £10,000 in just using, you know, my AI model to do basic tasks
1:43
like, you know, my admin, checking my account, making sure that everything is sort of lined up
1:47
Yeah, absolutely. And on top of that, a while ago, I actually took a libel action against a garage
1:52
which then, I'm not going to name the garage on for legal reasons, but I actually won the case
1:58
and they had a law firm with them. I obviously, I think one of the main things is I don't use AI as the final decision maker
2:04
I use it as sort of my PA, in essence, you know what I mean
2:08
So, Gabriel, you had no lawyer, you just had AI. Yeah, I had no lawyer, I just had AI
2:14
Wow. And I won. But, Gabriel, so this is what I mean
2:17
So I don't think I'm completely going round the bend about the need to reshore manufacturing
2:25
like physical jobs that you do with your hands that you make stuff that can be of course helped by tech The builder didn go out of business when they got the cement mixer But if we solely reliant on a service sector economy where you providing knowledge we done for
2:41
Absolutely, absolutely. I mean, some of the things I do, and there's some new functionalities I've just recently discovered with AI
2:46
So you can actually go into your browser, I don't know if you know it, and click your anthropic button, your cloud button or whatever
2:52
and it'll take control of your screen and actually do tasks for you. So I've actually used that
2:56
Yeah, so I've actually used that, for example, because I have a small wine bar. If I want to, for example, enter all of the wine lists, which we know we have 60-odd wines
3:05
I actually just upload the wine list into AI, open my screen, and it takes control of it and does it for me
3:10
without actually linking directly to the software. So I just give it the instruction, upload the wine list and say
3:15
insert all these wines with prices into my point of sale system
3:20
and just cross-check them to make sure that they're all accurate. And it does it for me while I can do other things
3:24
And this could have been somebody's job that you would have paid them full-time for
3:28
Absolutely. Unfortunately, I can't afford to pay someone else, otherwise I'd be doing it myself. I'm very small
3:32
So it would be something I would have to do myself, which would probably take me three or four hours
3:36
But instead of doing that, I can, you know, use AI and do another task at the same time
3:40
So I can actually do three or four tasks at the same time. So that's great. That's wonderful and amazing
3:45
That doesn't seem to me like it's going to end humanity anytime soon. Gabriel, really appreciate the call. Adam
3:51
Terminator vibes is what I'm getting, Tom. It is a bit, but I don't want to overhype it
3:56
I might go home and watch it and see if there's any similarities at the start
4:00
I wanted to just carry on with a couple of points from your previous callers
4:04
So the thing about him trying to make a picture of Donald Trump, I'm pretty sure ChatGVT has guidelines
4:11
For instance, I tried to make a funny picture. Let me, because it was provocative
4:17
So I think there's guidelines that it has to stay within and it won't let you do certain things
4:22
also the counting so if you ask it to count to 100 which i've done it will do it but you have it
4:29
will come back saying adam that will take a really long time and then i have to go i want you to do
4:33
it so you have to keep prompting it right to make it happen but i i wanted to paint it in a good light
4:39
and then finish on a negative so i'm re-qualifying as an electrician right now i've never been really
4:45
academic in school i just wanted to be the class clown and have a laugh in school but as i've got
4:50
older I've always wanted to explore where my mind can go but I have limits I have used chat
4:57
GBT for most of my course Tom and for instance algebra in school I never thought I would need algebra I started electrician training you need algebra I learned algebra in 10 minutes with ChatGBT after knowing nothing about it
5:12
You say you learned it, it makes it sound like you're now moving from Terminator to Matrix
5:17
so you just plugged yourself into the machine and you knew Kung Fu. No, so that's just me obviously in my brain, but I understood it better with 10 minutes of
5:25
ChatGBT as an adult as I did for my whole school career when I was younger
5:29
I see. Just understanding simple things and daily tasks as well. If I want some, normally I would Google something, I don't know
5:38
You were talking about Argentina a couple of days ago. I was listening to it on the way home
5:43
And I wanted to know, was Argentina in NATO? Was the Falklands part of the treaty
5:50
And it explained it all to me. Normally you would use Google for that question
5:53
but I just went straight to chat GBT and got it from there. Yesterday I was learning about sine waves part of my training
6:00
And it was teaching me about sine waves and stuff like this. For day-to-day tasks as an individual, I think it's brilliant
6:08
But just as humans, we always push things a little bit too far
6:13
especially if we want to turn it into a weapon. And I feel like it does give Terminator vibes and it could potentially go there
6:21
You know, you never really know. If you gave it the task, if you gave this incredible computer that can do mathematical problems in minutes that have taken us generations
6:33
if you gave it the task, let's say, of hacking into Britain's national air traffic control system
6:39
presumably it would keep going trying to do that until it did
6:45
Yeah, exactly. That's the scary part. That's the Cyberdyne Terminator vibes. that's when that area starts to come out
6:54
But then people are trying to do that all the time anyway. People are trying to do that. There's plenty of adversaries of the UK
7:00
who are trying to do that all the time. Yeah. But then it could get into the wrong person's hands
7:07
and that's when things can occur. Luckily, at most, I think, all nuclear missiles in the UK
7:16
they're on old stuff. They're not on new technology, so they can't be hacked like that, I'm pretty sure
7:22
I don't know, Adam asked ChatGPT. I wonder what answer it would give
7:28
Charles is in Hull. Charles. Hello, Tom, good to be on your show once more
7:34
Yes AI I got on to ChatGPT for two years and then all of a sudden it lied to me Yes
7:46
What was the lie? I just think it was really ironic. All I said was, I'd like you to make a cartoon with two guys looking at Donald Trump and saying blah, blah
8:01
and it kept having Donald Trump saying the words. And this went on for 20 minutes
8:11
It's like a constantly saying, you're not following my orders. And finally, after 20 minutes, he said
8:18
Chachy B said, yes, actually, I haven't been following your orders. I'm sorry about that
8:25
So why was it doing it? Because I can only think that the guardrails wouldn't let it
8:33
something to do with Donald Trump. But the crazy thing was, then I said
8:37
well, as you've been lying to me, it's a good job you're not in charge of the nuclear codes then, isn't it
8:44
And it said, yes, I suppose it is, really. Right. Great. But why are you talking to it
8:55
This happened six months ago. Yeah, but why are you talking to it like this? Just simply to make a cartoon of Donald Trump
9:02
I failed. It failed and I failed miserably. But if you're using it to make a cartoon of Donald Trump
9:08
this is not going to pose an existential risk for humanity, is it
9:12
You just turn it off and tell it's silly and turn the computer off. Sorry, it is because if it can lie, then there it's all, it can lie about anything
9:23
And it's already thinking. Was it lying or was it making a mistake? No, no, no. It wasn't a mistake. In fact, it was playing with me. I was its mouse and it was the cat and it was playing with me
9:36
And in response to you saying it's a good thing you don't have the nuclear codes, Charles, the way you made it sound, though, it was like ChatGPT had taken up the voice of Alan Bennett
9:44
It was saying, well, I suppose it's a good thing I don't. yeah yeah it was bizarre it was actually sort of rather sort of nomically sort of saying yes
9:56
it is a bit of a shame well you know i am doing this to you charles thank you well i mean there's
10:04
i see i would put that down as a mistake it glitched or something that would lie to you
10:09
that it would know that it was lying to you i think is a hell of a stretch isn't it
news