speleding

The Claude sandbox has a slightly different ENV compared to ChatGPT (LC_CTYPE set to C instead of UTF8 I believe) which upsets some scripts we have.

So in the CLAUDE.MD we tell it about that and then tell it continue reading AGENTS.MD. Not very hard to work around if CLAUDE.MD goes away, but just to support your point that the split did have its uses for some.

That balance probably depends on the human, and the context. If you are a beginner in a field the model should not assume you know what you are doing. On the other hand, for an expert it should try to work out what you mean with your vaguely worded order.

What I think should happen is that it should update its memory with notes on the proficiency level of the user, so it gets the balance right over time.

This is a problem if you allow your kids to use your ChatGPT account for homework (and silly pictures), like I do.

There are holding companies with zero employees. They typically pay a management fee to someone.

They now updated it. That bit is indeed gone.

If your app is not too large you can do CI/CD perfectly fine from your laptop. Put it in a docker container if you want to ensure it is perfectly reproducible. You still need Github as the origin repo to pull from, but the "git part" of GitHub seems to be (almost) always up and running. (We use "capistrano" for this)

Well, you would be surprised how many people think that blocking cookies means "no or fewer ads", even people in IT. No tracking, of course, just means it will show less relevant ads, not fewer of them.

So I'm not onboard with the "just block everything by default" crowd. If you frame the question as "Would you like ads to be more relevant to you" instead of "Do you want to allow tracking" you probably get a very different answer from users.

I would like the cookie banner to be changed to a browser setting, but I also would like the option to allow some sites to show relevant ads to me.

There is no need to be so fatalistic. Sure, AI will upset the apple cart, especially in our little corner, but the economy is built on pricing scarce resources and even if many things become abundant, most other things will remain scarce. There are only so many Manhattan penthouses or Rembrandts (dixit the Economist), so things of value remain, and it will be quite some time before a humanoid can fix the plumbing, so valuable labour will remain as well.

I, for one, am excited to see what the future holds. (side note: I live in a European country with a strong social safety net, I imagine that helps being unafraid)

I like the proposal, but I agree they could have sold it better.

This is basically a GET request that can have a body. I've found myself in need of that more than once when I did not want huge URLs with encoded data showing up in logs. Using POST request there is not appropriate because it signals data could be modified (i.e. cannot be sent to read-only instances). I guess modifying the spec to allow GET to have a body would pose too many problems.

The ideal solution for all this is would be what the Japanese have: a population trained to just throw garbage in the bin instead of on the ground. I don't think I've heard any politician who is even considering measures to get there. He would get my vote.

I agree that the government should do IT in house, ideally, because it's a core business for them. The reality is that it is very hard to attract and retain good IT staff on a government salary. The people who need to manage all that in a cost-effective way are especially hard to find.

So we end up with expensive consultants doing the work. Consultants have the wrong incentive. They don't want to stay in one place to long because it looks bad on their resume and overruns mean more money for them.

So really, I can see why a seasoned politician chooses the safest option for him. By the time an overrun occurs he will have moved on to the next job. I don't think left or right-wing politics has much to do with this dynamic. How will a left-wing politician magically get capable IT staff that higher paying industry can't even get enough of?

It's important to add the context that whenever our government tries to do something by themselves it ends up late and severely over budget.

So you have to weigh the risks of outsourcing to the risk of the whole thing becoming very late and very expensive. The risks around outsourcing are something further down the line, the risks of everything becoming expensive and late are something that will give the responsible politician a headache now.

The US-made Ultra UV light source is important, but there is quite a bit more than that to a $380 million lithography machine. (Otherwise someone would have created a competitor by now.)

Most economists think that tariffs are not a good way to collect tax, because it distorts incentives far more than e.g. a tax on wealth or property.

I run a low 7 figures SaaS as well. This is the blurb I answer with when asked about SOC2 (yes, yes, AI generated):

"While we follow industry best practices that align closely with the requirements of SOC2 and similar frameworks, we have chosen not to pursue formal certification at this time. Maintaining multiple certifications and undergoing recurring audits across the various regions in which we operate would significantly increase our operational costs and, consequently, the price of our service."

Text based accounting is a great use case for LLMs. I was pleasantly surprised how well Codex works with ledger CLI, especially in combination with git.

I wonder if this is going to give text based accounting a boost. Reviewing clearly worded git commits is so much more reassuring then letting an LLM drive your accounting package and hoping it doesn't mess up somewhere.

> ... something you know well and is easy to read ...

For this reason I tell my LLMs to use Ruby whenever possible. In one rare case where the performance of my script was critical, I told Claude to convert the working ruby script to Rust. It got it right in a single shot.

Getting Time Machine to work reliably over a network is painful, even the old Apple-made Airport with built in TM stopped working twice a year.

However, I have multiple Macs where I simply have a USB-C laptop SSD attached for Time Machine and they have worked without issue for years. These laptop SSDs come in huge sizes nowadays, and you don't need an especially performant one, so they can be pretty cheap.

I agree, but this app was created in 2006. That's when IE6 was still roaming the earth. Might have been easier to get it to work as an app.

It also works fine with Ruby and the "caxlsx" gem. Codex works fine with it as also.

Reading that BBC article, how the attacker got caught while shouting at an OpenAI building, it would seem likely that this attacker is confused or deranged. Not specifically someone with deliberate evil intent.

So the headline seems to be more "high profile person attacked by lunatic" than "OpenAI CEO attacked for being evil".

USB-C is decent for data transfer. It's pretty poor for power delivery: the pins are too close, so it's not rated for use in bathrooms or kitchens, and there are many more of them than needed for power delivery, making it relatively expensive to use in things like children's toys.

It was a mistake to conflate flexible power delivery and data transfer, you rarely need both at the same time. It's possible to design a better and cheaper 3 or 4 pin power delivery standard that can use higher power. But the law now says USB-C and good luck ever changing that.

Yeah, I worked on that one. It's passable, but I don't like the aspect ratio very much, it's too wide, I rather have 40" on 16:9

Agree! I still have several (now discontinued) Philips 40 inch monitors, and that is the perfect size to do programming work. Very little scrolling needed while you work. But I would love to have a 40 inch in 4K+ instead of 2560x1600, why is no one making these? (I did get a Samsung 8K 50 inch, but that's too large for a multi screen setup)

Are you saying democracy is working as intended, but you don't like the outcome?

Everybody made that exact same "slippery slope to Nazi Germany" argument when euthanasia was legalized here. That was decades ago. There have been several attempts to broaden or narrow the scope of those laws and the democratic institutions did just what they were designed to do, making changes judiciously.

If you are worried about the slippery slope, then you are really worried that democracy does not work as intended. (And depending on where you live that may be a very reasonable worry). By the way, Nazi Germany was not really a surveillance state, perhaps you are thinking of East Germany?

Well I agree, and my hopes aren't very high of this actually happening. Our politicians tend to be clueless with anything tech related, their opinions calibrated by what they saw in Hollywood movies, where anything tech related always turns into "black mirror". (By contrast, allegedly over half the Chinese politburo has an engineering degree of some kind).

But we could start small, with just one neighborhood, a pilot project where the kinks get worked out and slowly scaled up. Getting permission for a small scale pilot shouldn't be impossible.

I'm going to argue the other side: in Chinese cities like Chongqing they've seen a drastic reduction in crime after blanketing the city with cameras and monitoring technology.

Whole categories of crime disappeared. Women and elderly feel safe to walk the streets at night. No one locks their bike anymore in Chongqing.

I care about privacy, but I think we should be smart enough to work out a way to get some of those benefits without going full 1984. For example by having surveillance that can only be queried by an AI with very strong guard rails.

Admittedly, I live in a country with very strong democratic institutions, and I trust we would take action the moment something gets abused or surveillance overreaches. I would probably feel differently living elsewhere.

I'm not very familiar with deploy tools other than Capistrano, but I would think you also do not want to have the .git directory with your entire repo inside the working directory on the production server, so I assume some kind of "git export" must happen at some stage on most deploy tools? (Or perhaps they just rm -rf the .git directory?)

The article mentions .gitattributes but does not mention a super useful property you can put in that file: you can use it to specify that part of your repo should not end up on a production server. We have this line in our .gitattributes:

/test export-ignore

That means that when a "git export" happens from git to our production server it skips all test files. (In our case Capistrano does that, no additional configuration needed.) You never want test files on a production server and it saves disk space to boot. Normal usage is not affected, in development or testing you would always do a "git pull" or similar.

… and put local African cloth producers out of business. The same happened with shoes sent to African countries by NGOs. Well intentioned, but local shoe manufacturers went out of business. The local population did not really benefit, because traders would get a hold of the free shoes and sell them on for just a bit less than locally produced shoes.