Sean Carroll on limits of LLMs | Lex Fridman Podcast Clips

Lex Fridman Podcast full episode: • Sean Carroll: General ...
Please support this podcast by checking out our sponsors:
- HiddenLayer: hiddenlayer.com/lex
- Cloaked: cloaked.com/lex and use code LexPod to get 25% off
- Notion: notion.com/lex
- Shopify: shopify.com/lex to get $1 per month trial
- NetSuite: netsuite.com/lex to get free product tour
GUEST BIO:
Sean Carroll is a theoretical physicist, author, and host of Mindscape podcast.
PODCAST INFO:
Podcast website: lexfridman.com/podcast
Apple Podcasts: apple.co/2lwqZIr
Spotify: spoti.fi/2nEwCF8
RSS: lexfridman.com/feed/podcast/
Full episodes playlist: • Lex Fridman Podcast
Clips playlist: • Lex Fridman Podcast Clips
SOCIAL:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman
- Medium: / lexfridman
- Reddit: / lexfridman
- Support on Patreon: / lexfridman

[@LexClips]

Full podcast episode: ua-cam.com/video/tdv7r2JSokI/v-deo.html
Lex Fridman podcast channel: ua-cam.com/users/lexfridman
Guest bio: Sean Carroll is a theoretical physicist, author, and host of Mindscape podcast.

[@MarsAppSuite]

Lex, I know there’s a small chance you’ll see this but please minimize the overlap between these clips!!
It’s frustrating watching the same 5 minutes of the podcast under two different titles with just slightly different start and stop times.

[@josephhermanek5750]

We need to put these competing models of A.I. in a classroom and have them argue about bullshit abstract questions all day . . .
While we gradually die to entropy until we no longer have a stake in this topic.
🎉

[@unodos149]

I hope AI doesn't end up being a copy of humanity, cuz then it's gonna end up copying all of our bullshite too lol

[@HkFinn83]

lol!

[@ChrisPyle]

How could we ever program something we aren't able to be?

[@solution001]

Next is Mixture of Experts Transformers (MoE)

[@thatisabsolutelykooooge2211]

What’s that? Thanks ahead

[@caveman-cp9tq]

Intelligence is completely defined by input and output. The actual process is irrelevant as long as it works. If you think AI isn’t *actually* intelligent, then you will inevitably realize that neither are humans.

[@tylerlynn8225]

This is a gross oversimplification of intelligence

[@LegatusGaius]

I mean - what is your basis for saying this? It sounds wrong on its face. There is a very real difference between actually understanding a topic and simply probabilistically coming up with the “next right word” in response to a prompt to piece together a passable answer. As of now, LLMs will struggle exponentially as a conversation progresses and the amount of prior material that they have to analyze in forming their next response grows. I literally asked chat GPT-4 about this, so assuming its response wasn’t a hallucination, that is a real cap on the capabilities of LLMs with regard to the amount of compute required approaching infinity in a way that appears not to be the case with biological intelligence.

[@caveman-cp9tq]

@@LegatusGaius Does AI understand chess? Why does it consistently outperform humans?

[@caveman-cp9tq]

@@tylerlynn8225 Intelligence is just the ability to solve problems. If AI can solve problems then it is intelligent by definition.

[@LegatusGaius]

@@caveman-cp9tq the clip was talking about LLMs specifically. AI that are optimized for chess *may* (emphasis on may) have something like a conceptual model of how chess works, but it’s also possible that they are operating more or less probabilistically based on training data and wouldn’t recognize a chess board if you could visually show one to them. What is certain, though, is if they do have some actual understanding of chess - they absolutely don’t understand anything else. It is an extremely narrow form of AI versus something like LLMs which are significantly broader AI. I think it is almost certain that LLMs do not create anything resembling a mental model of what they are responding about, and are purely guessing “the next right word”. That is literally what chatGPT will tell you if you dig into this topic with it.

[@robertwynkoop7112]

Why ask him about AI, he has no real knowledge (just opinions), and being “woke” as he is his judgement and opinions should be questioned for authenticity as well as accuracy and validity

[@NamelessOne22]

"being woke" lmao.

[@jimbob8992]

Can you just for once tell us what you mean by woke? in context to Carroll. Because it just comes across as a nonspecific blanket dismissal of people who's 'opinions' you dont like.

[@yonaoisme]

people using the word woke unprovoked are not worth listening to.

[@garysherwood8902]

This. They don’t even know the answer but can’t admit.