Hassabis: 'We will need a handful of breakthroughs before we reach AGI' | Hacker News Reader

Hassabis: 'We will need a handful of breakthroughs before we reach AGI'

english.elpais.com

15 pointskaycebasques16 comments

I’ve sort of lost interest in AGI. It was always an interest because of all the cool things I imagined it would be able to do, but here we are now without AGI but with so many of the cool things I was imagining. So I care less if it's AGI or not. It’s extremely useful none the less. And as models get better and better and as tools improve we’ll see more and more value added to society even if it isn’t AGI.

I have developed ethical qualms about AGI. We end up either enslaving an intelligence, breeding it into submission, or end up with a powerful adversary.

Exactly. We are trying to develop something so human-like or more powerful so that it does our bidding. We're calling it "technology" and a "tool" but it becomes more than that at some point.

This. AGI-enthusiastics are disconnected from reality as they want something but don't grasp the consequences of having that said thing. Exactly like kids would want a dog because they only see they could play with it. But don't see the fact the dog needs to eat, go out for a walk, and sometimes go to the vet, can sometimes bite. AGI-enthousiastics are not ready to face an AGI, and no one of them is ready to take responsability for an AGI, given we'll see one in the next 50 years.

I'm still waiting to see any (net) value to society from our current AI. So far I believe it is negative despite being a paid ChatGPT user myself.

I was kind of hoping he would go on to speculate on what those breakthroughs might be, but the article doesn't go into any more detail. If someone like him doesn't have some next steps in mind, that doesn't bode well for AGI happening any time soon.

Just like fusion. But without the strong theoretical underpinnings so .. worse than fusion.

"Breakthrough" is the whiteboard covered in maths and an empty box labelled "magic here"

Humanity has known for decades that there's no known metal that will survive fusion's neutron output. They all turn into radioactive garbage.

Until we solve that, why bother with fusion?

From my admittedly limited knowledge gleaned from occasional PopSci articles posted here on HN over the years, don't they use plasma trapped in magnetic fields to contain the reaction and not a particular metal?

It's about the radiation, which isn't contained by the magnetic field. The latter "just" keeps the plasma from melting the walls.

Well, the magnetic or inertial confinement keeps the reactants dense/close enough to react.

The heat itself is entirely absorbed by the walls. As well as the previously mentioned neutron flux.

Also, heres one of many sources- this was found as part of looking up HFIR's name.

https://arpa-e.energy.gov/sites/default/files/B03-Wirth-Fusi...

Fast breeder reactors seem to survive fast neutrons.

It took like 60 years for the first big breakthrough, we need a handful.. see you in 300 years, if it’s linear

It might as well be 300 years, but looking at the logarithmic rate of technological development over the last 100 or so years, I can't help but think it'll be a lot faster than that.

if technological development is exponential while breakthroughs are logarithmic... the advancement is linear... so linear prediction is my best bet

That to me feels intuitively right.

Current AI will get better in the sense that it can solve phd level questions, but still does have the type of true intelligence a toddler has. There is some sort of spark missing there

Reply on news.ycombinator.com