nyssos

If it had nothing to do with the final result it wouldn't improve performance, and it does.

The relevant paragraph literally starts with "In India", where, yes, it's obviously the case that Hindus are not marginalized.

Be absolutely ruthless with technical debt. Opus is perfectly capable of producing idiomatic code in any mainstream language you please, but will seize on any opportunity to justify writing basically-python instead because that's "consistent" with the "convention". Deprive it of that excuse.

I also work on a large complex rust project (>1M LOC) with extensive use of Claude Code. It is very consistent with my experience. Claude frequently subverts the obvious intent of the system - whether that's expressed in comments or types - in the pursuit of "making the build green", as it so often puts it. It, like many junior engineers, has completely failed to internalize the lesson that type errors are useful information and not a bad thing to make go away as soon as possible. It is remarkably capable, but you cannot trust it to have good taste.

Yes, and? Something can be both scare and inadequate to a given task. FAANG L5s cost a pretty penny but I wouldn't trust a random one to prove a crypto library correct.

Agreed that we're not at saturation, but we don't have a canonical "best" either. For example ChatGPT 5.5 + Codex is, in my experience, vastly superior to Opus 4.7 + Claude Code at sufficiently well-specified Haskell, but equally vastly inferior at correctly inferring my intent. Deepseek may well have its own niche, though I haven't used it enough to guess what it might be.

You're presuming too much about what OP's quality standards are. Can SOTA models outperform the average junior engineer? Yes, obviously. Can they match the best human engineers, if those humans were given all the time and interest in the world? Equally obviously not.

I use hundreds of millions of tokens a month, and LLMs have completely transformed the way I work. They're also, frankly, pretty mid programmers.

I do this for debugging. Models are extremely vulnerable to framing effects and it's usually easier to spin up a fresh instance than it is to get an existing one to generate new hypotheses.

> The C that shows up in quantum mechanics is likely an example of this--it's a case of physics having a a circular symmetry embedded in it (the phase of the wave functions) and everyone getting attached to their favorite way of writing it

No, it really is C, not R^2. Consider product spaces, for example. C^2 ⊗ C^2 is C^4 = R^8, but R^4 ⊗ R^4 is R^16 - twice as large. So you get a ton of extra degrees of freedom with no physical meaning. You can quotient them out identifying physically equivalent states - but this is just the ordinary construction of the complex numbers as R^2/(x^2 + 1).

> but rather, physics uses C because C models the algebra of the thing physics is describing.

That's what C is: R^2, with extra algebraic structure.

Classical logic is the presumed default for mathematics, if someone is working in a different system they will say so explicitly.

Lawyer rates are extremely high variance, and the NYT is not hiring anywhere near the median. White shoe firms break $1000/hr routinely.

> I thought a partner at a large law firm might get $ 500k or more per year

Easily.

> but my understanding of copyright is that if a person when into a clean room and wrote an new article from scratch, without having read any NYT, that just so happened to be exactly the same as an existing NYT article, it would still be a copyright violation.

It would not be. Independent creation is a complete defense against copyright infringement.

Patents, however, do work this way.

Life expectancy at birth is irrelevant, people who die in infancy don't consume many resources at all. What matters is the percentage of the population that's of working age, and here Germany is fairly typical for a developed country.

Here's a less galactic version. Suppose you're implementing a binary tree where every leaf has to have the same height - a toy model of a self-balancing search tree.

Here's an implementation using GADTs and type-level addition

    data Node (level :: Nat) (a :: Type) where
        Leaf :: a -> Node Zero a
        Interior :: a -> (Node l a, Node l a) -> Node (l + 1) a

It's impossible to construct an unbalanced node, since `Interior` only takes two nodes of the same level, and every `Leaf` is of level 0.