Winsaucerer

This was a great read, thanks. Definitely provides evidence for my fears about vibe coded projects built by people who lack expertise to guide AI well. And if these are the issues you can see, it makes you wonder about what's hard to discover. Security to me is one of those things you want to be thinking about throughout the whole process of building it, not something you tack on at the end.

Some database tests can't be done within transactions. With postgres, I create a copy of the database via WITH TEMPLATE, and each test runs in its own copy of the database. Then it can use or avoid transactions as it pleases, because the whole thing is local to that one tests anyway.

That makes sense. The rest of us left ST a long while ago, and the rest remain because they're happy with it and have been for a long time.

I heard they are targeting Andromeda with the next round of tariffs.

I hate Imgur. Even with the app installed I find it doesn’t work well. I don’t understand why people use it — does it just work for them in a way it doesn’t for me, or are they more tolerant of its terrible usability?

What sort of code are you writing? I find a lot of my stuff requires careful design, refactoring an existing system to work in a new way.

If the code I was writing was, say, small websites all the time for different clients, I can see it being a big improvement. But iterating on a complex existing platform, I’m not so sure that AI will keep the system designed in a maintainable and good way.

But if your experience is with the same sort of code as me, then I may have to re evaluate my judgments.

I'm one of those people who thinks simultaneously that (a) current AI cannot replace developers, it just isn't good enough (and I don't think it's good for it to write much code), and (b) AI is simply an incredible invention and will go down as one of the top 5 or 10 in history.

I've said the same thing as you, that there is a LOT left to be done with current AI capabilities, and we've barely scratched the surface.

Presumably more questions can cover a wider variety of skills/techniques/topics. If the student doesn't know in advance which 10 will be selected, they either have to pray they're lucky, or work diligently on all problems.

I would love someone to do some research on attitudes towards how self-driving AI is, compared to factors like:

- Type of dev work (infra, frontend, backend, etc)

- Programming language

- Level of experience

- Quality expectations of project/work environment

The main problem I have with auto-suggestions is that they distract my flow of thinking. Suddenly, I go from thinking about my code carefully, to reviewing someone else's code. To the point where I get a bit stressed typing, worrying that if I go too slow, the suggestion will pop up. As you may guess, I therefore have them turned off :)

I am playing with Zed now though, and it has a "subtle" mode for suggestions which is great. When I explicitly want to see them, I press option key. Otherwise, I don't see them.

Is AI genuinely that good for you all? I can't leave it to its own devices, I have to review everything because (from experience) I don't trust it. I think it's an amazing technological advancement, perhaps will go down as one of the top 10 in the history of our species. But I can't just "fire and forget".

And that's not just because its output is often not the best, but also because by doing it myself it causes me to think deeply about the problem, come up with a better solution that considers edge cases. Furthermore, it gives me knowledge in my head about that project that helps me for the next change.

I see comments here where people seem to have eliminated almost all of their dev work, and it makes me wonder what I'm doing wrong.

I am also a fellow human.

Without having RTFA, I'd guess/predict that it will be possible to learn to only do this intentionally, much like we can think about raising our arm without actually raising it.

My own view is an idealist style one, where I think God impresses experiences upon us, and the experiences we have are determined by physical states. On this view, it's impossible to have a religious experience without there being appropriate physical states in place. In other words, agreeing with your conclusion.

There's a difference between reviewing code by developers you trust, and reviewing code by developers you don't trust or AI you don't trust.

Although tbh, even in the worse case I think I am still faster at reviewing than writing. The only difference is though, those reviews will never have had the same depth of thought and consideration as when I write the code myself. So reviews are quicker, but also less thorough/robust than writing for me.

This kind of thing is my main use, boilerplate stuff And for scripts that I don't care about -- e.g., if I need a quick bash script to do a once off task.

For harder problems, my experience is that it falls over, although I haven't been refining my LLM skills as much as some do. It seems that the bigger the project, the more it integrates with other things, the worse AI is. And moreover, for those tasks it's important for me or a human to do it because (a) we think about edge cases while we work through the problem intellectually, and (b) it gives us a deep understanding of the system.

Some people really do act like that, in my limited experience. Trump's opinion changes and lo and behold their beliefs have updated to the same view as Trump's.

I actually do think of him as a candidate for wealthiest person to have ever lived.

That's definitely a reasonable way to think about it. Another though is in terms of social status and ability to direct human labor, in which case most people are not more wealthy.

> we still have potholes to fix, t-shirts to fold and illnesses to cure

Only one of these things interests me. The hype of AI is threatening to kill something I actually enjoy doing. If the hype actualises, I'll likely find myself having to do something I don't enjoy. That being said, if programming can be automated, then probably every white collar job is under serious threat.

The title.

Interesting, I had a similar approach while studying philosophy. I much prefer to think and write than read, so I'd often think through a philosophical problem first, and then read others on the topic. I'd find that this order helped me grasp a topic well.

I don't know if I do the same thing with programming.

We had an experience not as bad as that, but cancelling on us a week or two before we were due to fly (and had booked everything else). They wanted US to cancel, but I saw no advantage to us in doing so. I assume it was to maintain some reputation on their side. We declined.

I've experimented (i.e., not used in anger) with doing something similar to this, except by placing the authorisation logic in the database via things like row level security and stored procedures. So if you run the query using the 'unsafe' method, then it skips security checks, but if you run it via the normal path, you're forced to provide details about the requester.

The difference here though is that the normal path only provides information about the requester, and not the actual business logic for whether they're authorised. That goes in the database in the form of RLS or other permission checks in stored procedures.

It uses a pattern similar to PostgREST, where there is a local value set with the user ID (e.g., 'set local "request.web.sub" = '{}''). Though there is some trickiness with connection sharing, to ensure that these settings get reset properly.

And to use an example from another commenter here, this means that if there's only 3 rows that should be returned, the database returns just 3 rows, rather than 10,000 that need to be discarded down to 3.

I haven't come up with a good answer regarding protecting which fields can be seen yet.

Sounds like the real challenge is experience/skill then? Easier to find people who know how to use AWS/Azure/GCP, than know either (a) your colo'ed setup, or (b) how to do a good appropriate modern colo setup.

(this is a question disguised as a statement, since I'm interested in your opinion)

There was a point where I started to dive more into specs, docs, and code, and less relying on finding the answer I'm after.

But in defence of those spending hours on forums, sometimes a project is not well documented, and the code isn't easy to read. In those situations though, I'll be browsing their Github issues or contacting them directly.

Often though I find it highly valuable to go read the docs, even if an LLM has given me a working example. Sometimes, I find better ways, warnings, or even information on unrelated things I want to do.

Timberborn also does this.

I like how scriptable it is.

This might be an ignorant question, but are these therapies likely to help chronic fatigue that predates Covid-19?

I'm genuinely not sure which one you think is worse (if any). (1) seems worse, but your reply suggests to me maybe you think (2) is worse.