DeepMind's Demis Hassabis on the future of AI | The TED Interview

[@TEDAudioCollective]

Listen to The TED Interview wherever you get your podcasts: bit.ly/3RTGROJ

[@LoisSharbel]

This young man is such an inspiration. I could listen to him all day. He's brilliant, charming and amazing.

[@GenericAF123]

I think Dr. Demis and his team is going to be the first to AGI. I think intelligence is what pulled us from the dark ages and I think enhanced intelligence is what propels us to the future.

[@marcabramsky1736]

Thank you so very much for bringing Demis on for a chat. He is absolutely brilliant. He reminds me so much of Ray Kurzweil in his positivity. I love listening to him. What a gift for humanity he is. I only wish I was intelligent enough to work with him. What is interesting in part of the training is using reinforcers to get a response from the AI or to build upon the knowledge base. The essence of Behavioral Science 101 is my own background. The SRC model in action. I wonder if there is an emotional response from the AI? Does it get gratification from the win or the right move? I wonder if they have any data around that? It is very hard to measure emotions but we generally do things for two reasons. The first is we seek pleasure. The second is we try to avoid pain. Really nice interview. I can't get enough of this wonderful fellow. Cheers!

[@pt20829]

What a brilliant mind. Thank you for this interview.

[@buckyzona]

how does this only have 12,000 views

[@kemalware4912]

It feels very wrong that almost no one listened these gold nuggets.

[@agaspversilia]

He and his team must win the Nobel prize

[@steve-real]

Demis is the most interesting human being on Earth. I have great hopes for his upcoming projects. He must have so many projects he wants to do. I would like to see DeepMind focus on biotechnology and fusion technology. I hate the term AI. It strikes me as phony.

[@JazevoAudiosurf]

games are made for training neural nets because you can generate infinite outputs for their inputs, the data is unlimited. the only other thing that is equally unlimited and complex is any real world process. I see in the long run that we shift from language models to AI that trains itself on visual and acoustic signals. autonomous vehicles are kind of doing that but they are not generalized enough. I expect something like the optimus bot to start walking through the world, observing it and mapping its actions to its sensory data. only then can AI really understand the world. perhaps the bot will have mapped images and sounds to words and by using an LLM understands the basics, and then only enhances that understanding

[@billyf3346]

the real test: can ai beat 'baldurs gate 2 shadow of amn' without any internet searching or guidebook cheating, knowing only basic english, basic math, and basic strategic thinking skills? if dm can build a general system that can beat bg2soa within a reasonable number of saves and play hours, without cheating in any way, then i will be impressed. starcraft has a clear objective, bg2 not so much. if these machines really have creativity, then they should be good at more non linear games, through learning, too. i really want to know if a learning system can solve something as challenging as baldurs gate 2 without any special training or internet knowledge. transferring learning over from starcraft would be fine though, as long as it doesnt use any internet or guidebook data. personally i really only feel mastering a game like bg2 from absolute 0 knowledge would truly prove that machines can truly learn, and in addition you would have the play record to pinpoint what it learned when, and i do mean it should be able to beat it in a human number of hours real gamers know bg2 is the real test. peace out.

[@lj9524]

Scary…AI We are going to be displaced by AI

[@GBlunted]

Did he say QBert lol?

[@lukesmith6130]

Does the King have a phone?

[@ramakrishna5480]

Oh man I just heard Sam.altman 1 hr audio