plorkyeran

With earlier models simply describing your desired end result often went poorly, and instead you had to break your prompt down into incremental steps. A single “sufficiently articulated“ prompt would result in the model forgetting about some of the things you told it to do by the time it got there.

Not any more, but it was common before Claude Code came along (~18 months ago). Based on the model versions used it may have still been the most common thing when the paper was started.

> We study a common intermediate setting: a developer pastes one complete file into a chat assistant [5], which sees the file’s internal structure but not its callers or dependencies. We evaluate the first answer, not an agent’s result after iterating on errors.

Very unsurprising results given this methodology. The agentic approach where models are allowed to iterate on changes is a very big part of what has made LLM coding tools actually useful.

My job is seeing a similar deluge of security work and >90% of the bugs have been in pre-2024 code, and zero of them have been bugs introduced after we were first officially allowed to use LLM codegen.

Invoking the Lindy Effect to proclaim that something that’s a year old will last forever is pretty funny. What the Lindy effect actually predicts is that something that’s a year old is pretty likely to only last one more year before being replaced.

I've seen swiping the CC used to buy the ticket be an option for things like printing the tags for checked bags on a flight, but it's definitely never been a requirement.

Hotels providing (extremely marked up) foreign currency exchange is a standard service in areas with lots of international travelers in most of the world.

It's the 500 most common words in that specific show rather than some broader corpus, so there's bound to be some words that happen to be unusually common or rare.

I have a 60hz monitor and a 144hz monitor and I simply can't tell the difference in normal usage. I can see a difference on something like http://testufo.com if I specifically look for it, and until I checked that I assumed the OS was just lying to me and the monitor was actually set to 60hz.

Your citations for “transplant” being used for illegal immigrants is people being told to go back to Ohio? None of your examples are about people who would typically be called immigrants, and as someone in coastal California I have never heard the term used for international movement.

I think you have very deeply misunderstood a pretty straightforward term.

If twilio does not want its numbers to be marked as spam then it needs to do a better job of not being used to send spam.

ISPs don’t give a fuck if their employees are casual pirates. It’d only be an issue if they think you’re going to run a seedbox or something.

It's a nice example of a prediction being so correct that it later starts to sound retroactively incorrect.

This is just a bizarre non-sequitor. Do you think that people are opposed to the death penalty because they’re worried that they personally will get executed?

Empty tags in XML are not implicitly closed. They require the explicit /.

Everything system-wide other than -march=native or equivalent has always been either placebo or sped thing up by introducing bugs. A specific software package sometimes didn’t build with appropriate flags and could benefit from tweaks, but having that do anything remotely meaningful is incredibly rare.

Claude code is an incredibly buggy mess. I have never used any other TUI that regularly has rendering errors or that is anywhere as laggy as Claude.

If your diet is exclusively protein and fiber then your maximum possible caloric intake is very low and how protein intake behaves when you are above maintenance levels is obviously not relevant to you.

> Were decomps/recomps a hobby because people wanted to make sure their old games could run on modern hardware and be extended with mods etc., or were they a hobby because people enjoyed the actual process of using Ghidra and reverse engineering by hand, and the decompiled game was just a side effect?

This is a very weird false dichotomy. It almost certainly was both of those for everyone who made any meaningful progress. You have to care about the end result to stick with a project to the end after you've finished the fun parts, and you have to enjoy most of the process of getting there for it to be a hobby. The journey and the destination matter, not one or the other.

There is indeed something special about getting to the top of a mountain and knowing that anyone you encounter there went through the same thing you did to get there.

All of our production Linux systems are running on Graviton these days, i.e. they’re arm64.

Well, at the minimum "Santa Clara gets cheap power because they host so many data centers" reverses the causality. SVP was already cheaper than PG&E before data centers existed at all, and the decent data center buildout hasn't pushed prices down. Santa Clara property taxes are typical for the area.

Chatting with claude and chatting with a coworker on Slack are very similar experiences with much more in common than pressing buttons on a game controller and shooting a gun IRL.

That’s the exact opposite of the work schedule I’d want if we were doing seasonal variations in working hours.

As opposed to the status quo, where the swatters are exploiting the police's behavior to commit crimes against the person?

He didn't quit/get fired/etc. His badge switched from saying Red Hat to saying IBM as part of an internal step towards integrating RH into IBM.

Do I understand correctly that GitHub currently doesn't support SHA-256 repos at all, so the author faked a screenshot of what the create repository UI would look like if they choose a really stupid way to add support for SHA-256 repos, and then proclaimed it an unsolveable problem? Selecting the repository format based on the first thing pushed to it is really not a crazy idea.

The submodule problem is real, but the forge problem is entirely just "if forges implement support in a way that makes it painful it will be painful", and that's true of literally every feature that requires forge support.

Because people get drugged and assaulted at Anthrocon a lot less often?

Apple does not ship any version of gmake. The only included make is BSD make.

Which is to say that it’s a very burdensome process if you think you should be able to go to Walmart, exchange cash for goods, and walk out with a gun, but pretty simple by the standards of regulated goods. The basic knowledge test is very basic, so the pain points are just that sometimes background checks are incorrect, the waiting period can be inconvenient, and just the general annoyances of government bureaucracy.