GPT-4o, AI Companions, and the Metaverse | Robert Wright & Paul Bloom

Рет қаралды 3,890

Ай бұрын

Subscribe to The Nonzero Newsletter at nonzero.substack.com
Exclusive Overtime discussion at: nonzero.substack.com/p/gpt-4o...
0:59 Bob’s sage status rises
3:54 GPT-4o: harbinger of the Metaverse?
13:29 Does sentience matter for AI companionship?
25:24 The promise and perils of social AI
33:22 How LLMs flip the script on computing
38:41 Heading to Overtime-and a special announcement!
Discussed in Overtime:
AI and porn (through an evo-psych lens).
Is sex passé?
Google’s new AI-boosted search; Bob’s and Paul’s SEO prowess.
Are the OpenAI resignations really a bad sign for AI safety?
Bob and Paul compare their p(doom)s.
Sci-fi TV to see (and skip).
Robert Wright (Nonzero, The Evolution of God, Why Buddhism Is True) and Paul Bloom (smallpotatoes.paulbloom.net, University of Toronto, The Sweet Spot, Against Empathy). Recorded May 20, 2024.
Twitter: / nonzeropods

Пікірлер: 29

@human_shaped Ай бұрын

Good chat. Some interesting, insightful and non-obvious points.

@Marty72 Ай бұрын

I’ve never used AI and thought it feels like talking to a friend. Even GPT-4o doesn’t come across as something I can chat to. Maybe it’s because I use it like an encyclopaedia, but nothing about it feels addictive.

@codica666 Ай бұрын

I feel the same. I tried one of these chat-AIs once, typed in a few questions and got bored out of my mind after 10 minutes. No clue how people get hooked to this stuff.

@MrBillythefisherman Ай бұрын

I think sentience will be achieved by a few things: 1) vision tokens are incorporated just as gpt4o has done with audio. 2) omni-modal prompts are trained on overnight 3) token limits are of the size that a whole day can be remembered, 4) multiple prompts are taken in and agentic behaviour is taken i.e if the battery is running out the model decides whether it should do the current task or switch to carrying out recharging. This is all relatively trivial stuff that I see solved in the next five years.

@baraka99 Ай бұрын

Great podcast if you know little about AI

@matttriano Ай бұрын

'AI' at the moment has a long way to go to become proper Limited AI; I have a feeling this conversation will be rendered moot dealing with the specific benefits and challenges of the thing as it is instead of the way we suppose it could be.

@tehdii Ай бұрын

38:00 can someone give me the link to this thought experiment Robert is talking about?

@Nonzero Ай бұрын

I was talking about this: nonzero.substack.com/p/chatgpt-is-capable-of-cognitive-empathy

@mmc577 Ай бұрын

I think AI therapists will be better in spite of their non-sentience. They will simply give the right answers you need to hear without judgment

@appipoo Ай бұрын

I.... don't think therapy is about giving answers or saying: "do this".

@sherrydionisio4306 Ай бұрын

IMO, the fact that perfection does not exist on earth with humans or with nature/environment, and knowing humans manufactured AI leads to the conclusion, AI will never be perfect. AI is learning on its own we’re told. I hope it learns that humans think much more though their emotions than critically.

@tehdii Ай бұрын

Rick Roderick did mention in his lectures how TV was supposed to alleviate loneliness, then cars, then internet... Sorry Paul, but you as a psychologist should know better ;) Kierkegard, Nietsche, Sarte wan not sf writers, they touched some of our deepest instincts ;)

@Michelle_Wellbeck Ай бұрын

Careful Bob on making the show too formulaic. In the future I might just ask GPT to make me a show where the host impersonates you just like this/

@MrBillythefisherman Ай бұрын

Massive reductionism: all it is a lookup table. You could describe the human brain as just a lookup table too. Its not and neither is an LLM.

@theoshouse8215 Ай бұрын

You guys are the best. I'd love for you to talk or write a bit about Biden's new tariffs. On the one hand, I'm skepitcal of protectionism in general. On the other, losing manufacturing jobs to China is one of the main reasons we got Trump.... I trust your takes on this.

@jennysteves Ай бұрын

No signs of plastics or PFAS etc for these future aliens to discover? Just how hard would these aliens look? Surely they’d be smarter than that .. smarter than these AI predictions, anyway.

@tylermoore4429 Ай бұрын

Too soon to take the demos on face value.

@mardomacfleno2474 Ай бұрын

I want to become a paid subscriber but I'm just a janitor with substance abuse issues. How much is it? How long are the videos? Okay, I looked. It's $5 a month if you buy the whole year. I still don't know how much more I get. I have to work 15 minutes to make that five bucks. It's going to have to be at least 15 minutes.

@larrydugan1441 23 күн бұрын

Think of AI as a toddler that has been taught to lie. This is what Silicon Valley is doing. Uncomfortable realities are not politically correct. What happens when the AI grows up. What will be the output? " Sorry Dave... I can't do that"!