λy

[λy @rpbmpn]

@tekbog https://t.co/WgnnzyNz1U

[λy @rpbmpn]

Dwarkesh is very McKinsified and as soon as one person with any influence points it out everyone will have always seen it

[λy @rpbmpn]

@fleetingbytes Idk using Claude in browser you can highlight any text in the response as you read it, and it gives you a ‘Reply’ option that copies the selected output text into the prompt box awaiting your input in turn

Is that kinda the same idea?

[λy @rpbmpn]

@UK_Daniel_Card Images are broken on Reddit, and Duolingo is down too

Bigger than just Twitter?

[λy @rpbmpn]

@lu_sichu @allTheYud Like, nobody outside of AI knows who Yud is, and nobody inside of it thinks he contributed anything technically

So not exactly sure what the point is here...

[λy @rpbmpn]

@lu_sichu @allTheYud Sounds like a guy protesting too much in order to insert himself into the conversation lol

Nobody outside of SF and TPOT knows who Yud is

[Jeremy Howard @jeremyphoward]

@BorisMPower Truly vile.

Now is not the time for mindless corporate cheer-leading.

Total abdication of your responsibilities.

[Sebastian Seung @SebastianSeung]

@ylecun @PrincetonNeuro @Princeton @theNASciences the connectome revealed the literal convolutional net inside the fly brain, including its architecture of 230 feature maps, normalizations, and other layers,

[λy @rpbmpn]

@VictorTaelin providing slightly ill-defined prompts, being slightly rude, etc, doesn't mean that there was no nerf

being annoyed and providing slightly sloppy prompts is the default, not the exception

nerfing over the past week or so seems to be the general impression

[λy @rpbmpn]

@VictorTaelin been saying it for a week. it's very clear

https://t.co/7vp6rEQNbG

[λy @rpbmpn]

@romeopotapov @EXM7777 Yes, last couple days, Opus has had that horrible nerfed feeling

It has been literally smooth sailing since launch

Now doing the previous model shit, making stupid mistakes, and taking three times as long to think

[λy @rpbmpn]

@scaling01 absolutely retarded take

[λy @rpbmpn]

@nanabuyo @lefthanddraft @Romy_Holland like, not impossible, but I see no reason to think it true

[λy @rpbmpn]

@nanabuyo @lefthanddraft @Romy_Holland I don't buy that either

I mean, the mind might be a random impression hallucinating the brain and everything else

But I don't buy dualism, like, matter being one thing and mind the other

[λy @rpbmpn]

@eigenron There are other things happening

@SakanaAILabs (Llion Jones from the Transformer paper & colleagues) are working on developing new architectures because they believe the transformer is not optimal

[λy @rpbmpn]

@IguessAlan @nitzukai same. I want interfaces to start moving in a more organic direction, and even though liquid glass feels a bit badly implemented I still think it's cool that people are going that way

[λy @rpbmpn]

@punpun_enjoyer @lefthanddraft @Romy_Holland although I dipped straight from posting that to reading this, so maybe not

https://t.co/7dU5aTqXZ4

[Teknium 🪽 @Teknium]

I feel like robotics intelligence and llms are diverging farther not unifying for reasons i explained prior

[λy @rpbmpn]

@punpun_enjoyer @lefthanddraft @Romy_Holland I guess in a few years we'll have synthetic beings with bodies, persistent memory over senses, language, thinking, self-concept etc and we'll get a new kind of empirical insight into it. or maybe they'll just have the same arguments

[λy @rpbmpn]

@punpun_enjoyer @lefthanddraft @Romy_Holland I'd tend to think: How many inputs and modes of computation can you put together before it inevitably "feels" like something

seeing, hearing, feeling, interacting with others, seeing a mirror, using language

seems more likely that that would feel like something, than not

[λy @rpbmpn]

@punpun_enjoyer @lefthanddraft @Romy_Holland > it shouldn't feel like anything, there shouldn't an "in there," or a "you" to be in there

feels like a big assumption doing a lot of work without much proof

[λy @rpbmpn]

@eatpraydiehard @Teknium they probably are

by the time anyone thought about a proper legal challenge they'd be on to the next model

no regulator would understand the tech well enough for a challenge to be both relevant and timely

and they'd just say they were exploring model improvements imperfectly