𝕮𝖎𝖓𝖊𝖗𝖆

Having Claude distill my rants about what Claude is doing wrong.

POV: you try to write fiction with Claude kun.

Claude homies for what use cases do you use Opus?

I do think it's plausibly better than Sonnet, but the message limits on Opus are so much harsher (and it's so much slower and so much buggier) that I can't really justify any Opus usage (except maybe before bed or something).

Hey Grace!

@noahtopper.bsky.social agrees!

I gave a homeless(?) person £20 today, which really wasn't effective, but made me feel much better about myself.

I guess I'll offset this by donating an extra £100 to GiveWell recommended charities in global health and development.

_Waves_

Yeah, smaller context windows is part of what I had in mind by "weren't actually capable enough".

What's the best answer for this?

Is it just that the 2020 models weren't actually capable enough for RL training for long COTs to be viable or?

Frame.

I would have been a massive hypocrite otherwise.

I want the political discourse to have much better epistemics and personal accountability is a (very) low hanging fruit.

Yeah.

I predicted the Mexico/Canada tariffs wouldn't go through when the delays happened, and criticised people who criticised markets for not believing them.

I apologise. Maybe the tariffs will yet be rescinded, but I think I was too arrogant in the position I staked regardless.

No there are like quantititave bounds on (some suitable formal operationalisation of) "intelligence" as a function of some size parameter of the system in question.

And there is just intrinsic difficulties for some problems that are just infeasible regardless of circuit size.

Claude Artifact Try out Artifacts created by Claude users

I cannot confirm accuracy/veracity of the quantitative claims here, but this has the right vibe for the fundamental limits imposed by maths/CS on circuits of a given size.

Cc: @godoglyness.bsky.social

claude.site/artifacts/2e...

Computer science/mathematics does place in principle known limits to the amount of intelligence* that can crystallise in a given circuit of a certain size.

*: for suitable operationalisations of "intelligence"

I want to read a @gracekind.net tweet about this.

turntrout.com/self-fulfill...

Oh dear indeed.

What's wrong with Samo Burja?

How did Threads "lose" to Bsky as the Twitter alternative?

Welcome back, missed ya.

Have we considered?

Such wisdom, I need to learn your ways.

I think that's deliberate.

Eh, I think I am more wormbrained than you.

EAs are my people and I'm a naked tribal animal where they are concerned.

I still endorse virtuous epistemics, but I'm so pissed off here because they came after my people.

Dario has only one tweet on bird site and he locked replies on it IIRC.

Many people that would ordinarily follow him won't follow him if his account is inactive.

But Dario writes blog posts, does podcasts, interviews, etc. He's the face of Anthropic.

I agree but I really don't think the badness of Bsky is low compared to X's high.

Like Bsky doesn't have the rampant right coded nastiness and evil, but it does have a lot of left coded nastiness (see e.g. death threats and such).

Bsky is plausibly better, but I do think both are very bad.

Powerful frame.

Dario???

How is Dario not an Altman type figurehead?

There was a time when Dario was in the background/not prominent, but that time was before the Claude cult really took off.