Are We On Path Towards Superhuman Intelligence? - Dario Amodei (Anthropic CEO)

[@LoFiLatentSpaceTools]

Glad for the re-upload since I’d missed the series, but FYI to folks like me this interview is from the Claude 2 Era.

[@drhxa]

You're right, although I suspect that by that time, they had already modeled what Claude 3 Opus model's capabilities would be like and the CEO, more than anyone, would want to know what that means in practice. Hence why his comments appear to be so relevant today. They're at the very cutting edge of this tech and it's super impressive

[@aazzrwadrf]

one of the best podcasts to date.

[@BartMassee]

love your interviews!❤ 🎉

[@deeplearningpartnership]

3:49 puts hair back in ear (for some reason?).

[@hannibal8049]

Hahahaha yes, its curious

[@alertbri]

Can you get a new interview Dwarkesh? 🤞

[@Megneous]

This is 7 months old. Get a new interview about Claude 3 Opus.

[@freedom_aint_free]

How about the 1.5 bit representation for LLM ? Surely it will accelerate things when implemented in real world LLM ?

[@ShawnFumo]

It definitely could make some kind of difference if it turns out the way it seems so far, and the big players aren't already using the technique. MS still hasn't released the code and I haven't seen any super serious replications yet, but I'm guessing there will be attempts soon. Check out the announcement from Extropic today on another interesting direction things may go (harnessing the inherent randomness in analog systems for more efficient hardware).

[@quantumspark343]

2-3 years?! His model claude is like that already and gpt 4 too, i dont get it

[@Bassoarno]

It can't drive a car

[@vishalmishra3046]

Seems like old video. Please add the video recording date to your video's description. Thanks.

[@0113Naruto]

We don’t understand intelligence…

[@Apjooz]

It's an advantage.

[@JaseyJ]

Shutout

[@Capefear83]

Mini cuts are mid