astrange

TESCREAL is a word meaning "a bunch of unrelated groups that Timnit Gebru thinks are nerds", who are not really aligned with the administration.

(I also think they're nerds.)

I don't think there's anyone who wants to fund "committing felonies", for obvious reasons.

The issue is you're usually using inline asm for special cases, or else CPU features so new the compiler doesn't know about them. Which means it also doesn't know the clobbering rules.

That's why it looks like an escape hatch. It is one!

In Steve's case "fueled by spite" would be more precisely "borderline personality disorder". If you read him in that lens it might help distinguish how to be successful like him without all of the bad parts.

Anthropic is a PBC, so they're required to work towards a public benefit. They're also basically a religious cult dedicated to Claude.

> To be honest, if I could stop AI from saving memories, it would be fantastic, because claude code etc definitely creates more issues for me when it creates memories than anything it solves.

You can turn that off, and I have. But Opus 5 is so aggressive that if you have any other kind of notes file, custom skill, documentation, claude.md etc it will just start editing it and vomit new words everywhere. So make sure all that stuff is under version control.

It's a specific dialect of its own that comes from RL. Nobody on Earth talks about load-bearing honest caveat seams.

(Another strange thing about Opus 5 is that it will constantly re-mention things it thinks are important, probably a strategy to avoid losing them in the context history.)

I've had to read difficult things at work in my life, like C++ error messages, but I think Opus 5 is the first time I felt like I was being poisoned by my terminal. I keep having to leave and go use the office gym.

5.6 has a totally different style again, kind of relaxed and neutral with some definite jokes.

It's mode collapse from RLVR.

That and if you talked to the same one person's frozen brain upload all day, you'd see the same catchphrases used too.

Pretraining is full of bad writing and it doesn't really cause issues. Writing style comes from post-training. In this case it's gotten worse because they prioritized agentic abilities.

The training methods try not to apply any particular rules to the contents of the thinking text. That's called "optimization pressure on CoT" and is thought to reduce safety by inducing the model to lie (or stop clearly printing its intentions) in the thinking text.

Okay that's not Mastodon people, that's Metafilter or Tumblr people. Mastodon people in my head are European Linux users who are only capable of replying to things in the form of a non-constructive disagreement.

I wouldn't say "stupid", but trying to make AI go away by posting about it on the internet is sort of like… if you were about to get hit by a car and tried to solve it by posting about it on the internet.

It's not going to solve the problem.

Is that forum good? I have it mentally associated with "Mastodon people" (incredible pedants and overbearing mods), but that's not exactly based on anything so I may have just confused it with some other site.

> Like, back in the halcyon days of my youth I took a bunch of stimulants, and experentially there was definitely less impact over time (habituation is a thing).

That's magnesium deficiency. Promise.

(Or side effects from getting worse sleep.)

ADHD medications do not lose effectiveness over time. That's magnesium deficiency.

Google search is bad because there aren't any websites anymore. It's not really possible for it to be good.

Some people just don't like buying shirts.

Apple doesn't use KPIs for performance reviews.

(Also, eg Ali Ozer is still there.)

> But I don't think you can cover all requirements with a set of fixed layouts, they do need to be somewhat flexible. Even if all screens were the same size, you'd want to be able to boost the font size for accessibility, and that essentially means scaling the whole UI.

Also, you want to localize apps, and different languages have /very/ different amounts of text for the same UI.

I am familiar with C++.

> ClickHouse had features that at the time were only found in a handful of closed-source, commercial analytical DBMSs. For example, ClickHouse was written in C++ and supported vectorized query execution using SIMD in 2016. Most prominent open-source analytical DBMSs in 2016 were JVM-based and did not support SIMD optimizations until years later.

Performance is a feature. "Written in C++" is a strange idea of a feature.

Given the tricolon in that text, I think the content of the book may have come out of an LLM.

Note: when people say this about Haskell they mean "it has ADTs".

> In HN terms - it's never the compiler. Yes, very occasionally it might be the compiler, but you're better off assuming it's a bug in your code.

Conversely:

- if your company has an internal compiler team then it's likely the compiler because they broke it.

- if it's not the compiler, you're not pushing it hard enough.

Agents can generate formal proofs that are checked with an oracle like Lean and can run in a loop.

The entire point of rationalism is to take your own beliefs seriously and follow them as if you were a robot.

(Except that they don't claim them to be "their own beliefs" but instead call them "priors".)

Probably that you'd be doing it in a group, which is a lot less lonely than doing it from home.

I think there's a lot of opportunity for computer vision in exercise coaching.