News for hackers. Stuff that matters.
Re:DeepMind says reinforcement learning is ‘enough’ to reach... by bgroat on Thursday June 10, 2021 @03:05PM (#27464318)
I'm not a neuroscientist, an AI specialist, or a hardware engineer.

But as an enthusiast of all three I really think that AGI is a hardware problem, not a software problem.

Reinforcement learning on a massive corpus of data is how we train all biological intelligence.

The crazy thing is that in humans we manage to do it on ~3 watts an hour.

I think we have the software cracked, my gut thinks silicon just isn't the right material

Re:DeepMind says reinforcement learning is ‘enough’ to reach... by pkAbstract on Thursday June 10, 2021 @03:21PM (#27464505)
You may be right, but it's also commonly believed in these communities that hardware is the part that's already been solved. Computer hardware already vastly outstrips human capacity in many domains.

To me, it seems more likely that we're missing something/some things on the software side. AGI could probably run on present day hardware or even older.

Re:DeepMind says reinforcement learning is ‘enough’ to reach... by 6gvONxR4sf7o on Thursday June 10, 2021 @03:42PM (#27464772)
Silicon is likely fine as a material. GPU cost per operation is still dropping insanely quickly. A lot of really hard ML problems are just making big things feasible, or sampling big things enough to get decently precise estimates. With 10x the GPU power and memory, a lot of this gets easy. With 100x, some hard things get trivial. At the end of the day, GPUs and TPUs drive AI research more than anything else as models grow massively.