giraffe_lady

Unfortunately apple accomplished their goal of killing that entire idea. The time for it to take off was 10 years ago. They begrudgingly support it now but it's too late. gg

If you keep them distinct unification is a weapon to be used against you. You're writing your own blackmail someone just has to call you on it.

With prose fingerprinting, sophisticated tracking, now your identities are only separate by rapidly eroding social convention. Intentionally merging them allows you to have control over the process, and helps you maintain discipline about what you reveal where. If you don't do it it will be done to you.

Commit messages now do double duty: log of the changes, but also signing off that you've read and understand the code, are taking responsibility for it under your own name.

The second part is an emerging convention but it's fairly common from what I've seen. IDK maybe no one will be reading or writing code in 18 months but if they are we'll need to separate these two things.

Secondarily but probably more importantly: it's real easy to read something and say "yep I understand that." It's a lot harder to fool yourself when you have to actually write down, with your own brain, what that thing you read does. If you're not doing this you must have incredible discipline elsewhere to make up for it.

I'm pretty sure that's the general trend and it will continue.

But I do think what benefits LLMs is the speed and accuracy of feedback. Type systems cover the accuracy part, but haskell was killing them on speed. It seems like a strange choice to go so far the other way on accuracy when there's a lot of languages in between. But I'm not familiar with the project so not in a position to call it.

It's not also really about expressiveness IMO. I've found LLMs to be best with more constrained type systems: they are better at ocaml than they are at typescript.

It sounds neat when you phrase it that way but my decade of professional experience in typescript has not shown it to be the case in practice. ymmv I guess.

Algorithm W is 40 years old and HM has vastly better ergonomics than TS's types.

Why certain approaches didn't catch on until recently, or ever, is an interesting thing to think about but "we didn't know how" is not the story here.

Algorithm W is like undergrad level of sophistication. People who like HM more (and I am one) don't like it because it's "advanced" and to some extent exactly because it isn't. It's sound and fast and infers almost everything. TS seems to have one of those features now, so that's nice.

I don't think I learned this skill from the sat.

I was self taught and went back for a bachelor's after ten years working and then a masters after another four. There was a lot of wasted time in the BS (ie the trade-school-for-programmers classes) but the pure CS has been valuable to me quite regularly as a working developer.

Not every day, maybe not even every month, but I've faced plenty of problems I was well equipped to solve directly because of the formal study of them.

A lot of times it is simply following some geographic boundary that is obvious locally but not on a map like this: a river or whatever. Even if the boundary isn't itself a major obstacle now the settlement history is embedded in the social geography.

Or like look at that corner of northern indiana. That's the part of indiana that is part of the chicago metro area. Makes absolute sense to avoid cross-timezone commutes.

Northern michigan is pretty damn far north! Winter sunsets up there would be before 4pm on central time. Makes sense for them to shift it permanently.

I think they probably all make sense if you know how to look at them right.

Font shaping rules are actually turing complete. An incredible amount of engineering effort has gone into this not being a total disaster everywhere at all times.

It's more like asking what editor and keyboard layout they use. Highly relevant to the user but you should simply assume someone describing work is using a setup for it they find productive. If you decide to dismiss their output it wouldn't be over these details.

> Policing, especially in the United States, has developed a reputation for cruelty in the last decade or two.

Kris Kristofferson was writing songs about the petty cruelty of cops in 1970. Fuck tha Police was released almost 40 years ago. Richard Wright published a novel in 1941 about a man who was tortured into a false confession by police.

The technologies are new and powerful but that's not the issue. The issue is that police are themselves lawless scum who will use any tool at their disposal for whatever harm they feel like inflicting, to who they always inflict it on. We should stop giving them new tools, absolutely. But we also have to stop them.

I’ve done it over less.

Probably coulda used an example that isn't itself a fourth amendment violation that essentially requires perjury to accomplish. Also less euphemistically called evidence laundering. Not really a neutral example.

I had to get into an old tcl program for work recently and had the same thought. I wouldn't necessarily pick it today but it was kind of nice in a way that's unfamiliar to me from modern development.

It assumes $44/kg to launch into orbit. Idk where that came from if it's a musk claim or what but it's ludicrous against current costs. Currently still at thousands/kg and dropping but nowhere near $44 and I haven't heard anyone claim that's likely who isn't selling a rocket company.

> Many will tell you a story about how their family desperately escaped on a row boat, are lucky to be alive and blame Castro.

Many will also not want to tell you any stories about what their family did in cuba.

Like I'm not trying to downplay the real violence and excesses. But read about how americans talked about cuba pre-castro, what they expected from the place during batista's dictatorship.

The deeply entrenched feudal-type rural poverty provided the raw human material for the mafia to create a vile playground for their own fantasies and sell it to other wealthy americans. Epstein economy shit. Castro didn't come out of nowhere, he harnessed the deep disgust and pain of a people being treated as things for money and pleasure. So yeah the sugar plantation landlords, the batista enforcers, the mafia sex slave sourcers were running for their lives. A lot other people got caught up too, absolutely. But ask your hypothetical cuban immigrant some followup questions. It's not always exactly personal why they hate castro.

Bullshit I just pasted random shit from google in until it worked and then instantly forgot which combination of the 20 things I tried got it there.

We're hoping to have a preprint ready at the end of the month, I'll send it to you then!

> 2. Do we actually have any data or evaluations that show which typing discipline is better for agents? The only benchmark I am aware of [AutoCodeBenchmark] has Elixir come first (dynamic) and C# as second (static), so it doesn't answer the question. There are other benchmarks that show dynamic languages require fewer tokens to solve problems (but that's not a metric I particularly care about)

I am actually writing a paper on this right now so nothing I can point you to yet but yes. LLMs are better (produce working code in fewer attempts controlling for the relative size of training corpus) when using type systems with inference and global unification. It is largely about the quality of the error feedback channel so languages with very good compiler errors (accurate, localized, include the correction with the failure) can close a lot of ground.

But inference + sound type system gives you a constraint propagation that genuinely restricts the ability of the LLM to get into trouble. Type systems that require annotation give up most of the benefit, since the annotations are themselves surface area for LLM mistakes. Unification also puts heavy limits on the expressiveness of the language which is a confounder and may actually be a big part of the benefit too.

Everyone has been on the "the training data is better" thing but I actually don't think so. All of the languages that people report as being better because of good training data actually have fairly restrictive type systems. Elixir is an exception, but it has exceptionally good error messages! And also, along with erlang, pretty unique runtime semantics that may contribute but that's outside my domain I'm on type systems. Debunking the training quality thing is not what I'm working on but I have deep suspicions about that common wisdom.

> Is anyone collecting and writing on how to understand this architectural style?

Are we on the same site? Is anyone writing about anything else?

I actually don't see any personal productivity improvements from using opus over sonnet for coding. If you're keeping tasks small and conversations short, reading the code and correcting before changes go in, whatever advantages opus has aren't practically significant. It's also just talky as hell, overexplains anything it touches and every token produced this way increases the surface area for hallucination so you need to have your guard up even more with it.

There's a sweet spot of complexity for low importance tasks where it's just big enough I don't want to do it and just simple enough to have opus plan/delegate/review with another model. So possibly model improvements will grow this window, but currently I don't do much in there.

Anthropomorphizing is a shorthand for a powerful and poorly defined set of metaphors. There are tradeoffs going both ways but trying to dismiss it as merely "strange fixation" shows your own weakness.

> "Can you kindly consider the following problem" not how anyone would actually speak to a valued collegue one considers smart.

Man idk, it's not how I talk but there's like 100 million nigerian english speakers, twice that indian, and they have some speech mannerisms that surprise me the first few times. I'm pretty sure I've heard exactly this from a colleague before.

Intuition about what a native speaker would do with english are scrambled right now. I'm not even sure most english is spoken by native speakers anymore, and the boundary between a native speaker and someone who has "merely" been using it as their educational and professional language for their entire life is disorienting.

I'm not sure if this is in the anthropic models themselves, or just the harness, but they can self-initiate ending the conversation and reportedly do it if you're using abusive language towards them.

Don't need to wait that long the inevitable data breach will be bad enough.

There's a lot of french in tolstoy too, and I believe most of the dialog can be inferred to be in french but is in russian for literary reasons. French was the daily home language of the russian imperial aristocracy. They spoke russian with an accent.

Sorry you're right, I should have been more careful with the phrasing there. The structures, risks, and incentives that make business owners believe and act like this are real, and that's the poison I was pointing at.

I'm not trying to put all of the blame on individuals responding to the pressures being applied to them. But neither am I accepting their abdication of the responsibility to act with honor and courage in the face of those pressures.

> Hard-working founders should be able to cash out

Why? Operating a successful business should be remunerative on its own, or else it's not successful. Owners who don't want to do it anymore can let it become worker owned. If they don't want it, it can dissolve. What else do you need? The very concept that the end of a successful business is a big payday for its creator is itself the poison here. There is no end just another workday, success is ongoing not final. This is natural and correct.