I really can’t sit through 2 hour podcast about this .
Can you give me a brief summary and your views ?
Should we genuinely be frightened? Is it really plausible that AI is going to change the world possibly in the next 5-10 years ?
You really should. In terms of importance, climate change is nothing compared to this.
And btw, the short answer to your question is yes.
There's a few things that have caused very senior and influential people such as Geoff Hinton to start to worry. One is the fact that these models already learn way better than the human brain does. He and others invented them in the first place to better understand the workings of the human brain and never for one moment thought that what they were creating would become better at learning than the brain is. The second scary aspect is the extent to which he believes (as I do) that these models do already think. Not in the same way we do because their architecture is completely different from our brains. But nevertheless they do understand complex problems and can reflect and hypothesize just like we do, already. And the third thing is that they can replicate their knowledge between them almost instantly, whereas it takes years for humans to learn advanced concept. When an AI model learns something, it can pass that learning to another instance immediately.
So these large language models have already developed capabilities we never dreamed they would have, and have done so at a rate not many imagined. The consensus now is that these things becoming more intelligent than humans is not so much a possibility, but more of an inevitability. And once that happens, who knows what the outcome may be with "beings" which are smarter than us, woven into the fabric of everything we do.
GPT4 is already more knowledgeable - by orders of magnitude - than anyone on the planet, drawing as it can on the entire content of Wikipedia and other sources for it's information. And as we feed these models with more and more data - video this year - they are becoming more capable and able to reason even better than they can now.
The very immediate danger is not that they will turn all "Skynet" on us however - though they may do in time, no-one can be sure that they will not. It's that bad actors - terrorists, governments, individuals - will use these new superintelligences to do bad things. Like develop and release new nerve agents or biological weapons for example.
That's aside from the changes coming with entire industries being wiped out with all the employees being replaced by bots.
The world is going to change radically and very very quickly. We are not talking decades away, we are talking years.
www.bbc.co.uk