The AI Field is Acting Crazy - Demis Hassabis (Google DeepMind CEO)

Full Episode: • Demis Hassabis - Scali...
Transcript: www.dwarkeshpatel.com/p/demis...
Apple Podcasts: podcasts.apple.com/us/podcast...
Spotify: open.spotify.com/episode/6SWb...
Follow me on Twitter: / dwarkesh_sp

[@andresgomez7264]

Very smart and down to earth guy. That's a rare combination these days

[@hanslick3375]

Very smart? Contender for the greatest genius alive.

[@StamatisSkiadas]

Horse population suddenly reduced by 90% due to adoption of the automobile. Horses were not required so the biomass reduced by 90%

[@richardnunziata3221]

Seriously the focus should be on creating flexible secure infrastructure interfaces and institutions that are resilient to automation take over. Unfortunatly they will not change until they are forced to change . That means open development of AI systems is the only way to force institutional response which is already by their nature sluggish , outdated, and burdened with bureaucracy and special interest (regulatory capture)

[@Megametalwolf]

Dr mike! What happened bro

[@yang8244]

What i hear is "we made it, we are a top dog in AI, now lets be cautious", aka, lets bring in government regulations to make life harder for upstarts.
after all, "unsupervised' competition is dangerous...
The last thing you want is for governments to have their paws on innovative ground breaking technology but obviously corporations are interested in government involvement. Much easier to sway a few politicians to pass some regulations than to persuade consumers to not go to the competition.

[@SigmoidalHive]

no ulterior agenda behind this at all...

[@Phasma6969]

Pretty much yeah 😂😂

[@squamish4244]

Not with him. I'm way more concerned with what Jeff Bozo or Mark Zucky will do with AI.
It's hard to argue anything bad about AI's potential in medicine and materials science. In 2050, twice as many people will be over 65 as today, which will be a nightmare for healthcare costs unless we can cure or ameliorate most diseases.

[@SigmoidalHive]

@@squamish4244 it's kind of odd that you'd make an exception when the rest of us are saying anyone dictating or manipulating what an entire field can or cannot do is untrustworthy.

[@noone-ld7pt]

​@@SigmoidalHive I personally try to judge these guys on their actions over their words and I've found the models deepmind has released openly (Alphafold, Graphcast) to be the most inspiring in terms of the realistic scientific potential of AI.
Sure it inevitably feeds back into googles agenda, but if that means massive openly available scientific breakthroughs then that doesn't seem like such a bad agenda to me tbh

[@SigmoidalHive]

@@noone-ld7pt in terms of actions, alphafold & graphcast would be significant for an independent research team. do I weigh these at the same level of public good as, say, Folding@Home? debatable.
but let's be honest, these demos do not have the same scale of impact as, say, the ML in UA-cam's algorithms.
don't forget, Google Deepmind also blogs about even more impactful, less transparent research like the materials synthesis project. maybe the average layperson doesn't pay attention to how these are received, but I can confidently say it leaves me less trusting of Google.

[@findmeinthecarpet]

I need AI-enforced DEI in my life already! Hurry up!

[@yosup125]

for the algo

[@stephenthumb2912]

Which is safer Linux or windows. He unintentionally is advocating for gov controlled AI and crushing open source. He should quit google and work in open source.

[@darylallen2485]

How would someone fund the compute resources needed for modern ai in an open source structure?
Hint: OpenAI tried what you're advocating already and couldn't raise the capital to do the work.

[@BlimeyMCOC]

There’s Meta I guess

[@user-yl7kl7sl1g]

Move fast and break things, is the correct approach. You can only test a tiny sliver of possibilities in simulation and behind closed doors. Yes, people will die by releasing early. But the alternative of taking five years to build something "safe" then releasing it the the public all at once, only to find out there's some situations that cause mass casualties or worse, extinction. You need to increase Ai gradually, with maximum users, while improving alignment continuously.

[@saeedsahal2098]

Algo

[@andybaldman]

Demis is the only person who has done anything useful with ML. Everything else is overhyped chatbots.

[@Phasma6969]

They are a whole team and not just a singular Messiah

[@eldoprano]

mimimimimimi

[@TheManinBlack9054]

True, in terms of actual scientific achievements Deep mind are the people you're looking for

[@andybaldman]

@@Phasma6969 I never said he was a messiah. I was speaking for him and the teams he has lead.

[@Phasma6969]

@@andybaldman fair enough mate 🐥