This feels like a really good blog post to write - something I'd really want to read - but the generated AI prose distracts, and makes me lose trust and not want to read this. I'm very AI-pilled, so this is not a criticism of AI as a whole, just this piece.
This feels like the sort of decision that well meaning folks that manage utilities can disagree on. It's frustrating every time it happens because it's usually a fairly reasonable perspective to ask for XDG / operating system specific support for folders to be respected. (e.g. macOS's ~/Library stuff).
Posting a github issue like this though feels like an easy brigading target, so please don't do that.
https://hckrnews.com/ most days, has a good line which tracks what you saw last. Sometimes filter by top 10/20 when I've been more busy and to catch things I may have glossed over as being more popular.
This, but I'd say that there is an engineering perspective that you can apply. What the developer world is trying to discover here is how to encode taste concretely - often this is things like really mechanical design system things (space/padding/margins/relative sizes/color contrast/...).
The hard point is doing that well deterministically in a way that is token efficient and which can be iterated on. I suspect it's difficult to encode that into an AI model of what's good taste.
I've done VHS (terminal recording) rewrite in rust built on libghostty called Betamax [1], and then threw it at creating a TUI for jujutsu. This worked pretty well, but is admittedly a much simpler task than a web ui. The harness was able to build out a feature, confirm it worked, find bugs in what it expected vs saw and fix them with evidence. Web is many more pixels and more nuanced things that have to do with how various elements relate.
I speculate that maybe the OpenAI models have the edge here over Claude's (while still being terrible at web UIs), as the training necessary to handle computer use stuff is heavily image processing based. I wouldn't bet on either being bad for much longer though.
Some agents have specific tools that the models have been trained to use. E.g. diff formats for editing that aren't the same as the standard unified diff format. Access to specific thread / subagent / etc. tooling or the base prompt can perhaps also impact how the tasks are completed.
Yeah, what bothers me is that the prompt already said using a different vulnerability didn’t count, and the model did it anyway. We’re starting to assume clear instructions act as real constraints, but here the measurable goal seems to have won out and the rest became flexible. That gets pretty worrying once the agent has enough capability and access to find its own shortcuts.
Clear prompts have never worked as real constraints. Ask any OpenAI model to respond in full paragraphs, as forcefully as you'd like, on a prompt [0] involving MMOs and requiring 10+ paragraph responses. The middle will be three-words-per-line drivel, with seemingly no way to avoid it. The exact way in which models deviate from instruction changes from time to time, but they're not "aligned."
[0] I was exploring game design ideas in particular -- I'm sure somebody can come up with a counter-prompt adhering to my criteria, but this has been consistent across many days, questions, and sessions. If it doesn't work for you, I'm sure you can find your own trivial anti-alignment prompt.
reply