@baldur “We believe that quality is more important than just piling on features; we believe that quality is the most important feature. And we believe that high quality is transformative — it makes for an app you never hesitate to reach for. You can rely on it, and you do, again and again.”
This is precisely the principle that is violated when someone sloppily gorilla-tapes LLM features onto their app for no good reason.
@baldur right. Such a lengthy regex necessarily has a lot of marginal cases that need to be verified or at least tested. But the state of practice being what it is, I imagine the developer wrote (at best) some random accept-reject tests, and trusted that the LLM verified the regex.
Which is asking a lot of something that demonstrably lacks mathematical reasoning but oh well.
Making that verification cognizable to developers is a very hard problem which of course LLMs don't solve; it's new.
Also, I’m the person on the private mailing list who warned people about Trollope’s racism. The reason I I used the toned-down wording I did is I knew I was in an audience of Trollope fans who are almost certainly blind to how much of a reactionary classist he was.
Unless you have experience teaching or training a variety of web tech (HTML, JS, CSS, SVG, etc), you likely don't fully understand their relative learning difficulty
Basing your assessment of which is harder on your attempts to teach yourself is esp unreliable
It's hard for you to know if a tech is hard or if you were just unlucky in stumbling into a bad entry point. Getting into something with the wrong mental model means you have to first unlearn a bunch of crap before you can actually learn
If you're unlucky enough to start at something with the wrong mental model, it's always going to be a struggle until you actually do the work to unlearn it. This is very hard to do on your own because we generally aren't aware of our own worldviews
And comparing the relative effectiveness of tech you don't fully understand versus one you do is always going to favour the latter and does not accurately reflect the capabilities of either
“WebAssembly: A promising technology that is quietly being enshitified”
"So we have this WASI thing that is only version 0.2, and that is already way more complex than the alternative solutions. Is there any way that it will succeed?" https://kerkour.com/webassembly-wasi-preview2
@baldur Its worth emphasizing: These skills don't goes out-of-date! Browserdevs go out of our way to ensure the pages you've written yesterdecade still work!
New optional features have been added, & best practices have solidified since the 1990s... But still!
Having to periodically rewrite outdated code is something webdevs bring upon themselves! Having to keep ontop of the latest frameworks is something webdevs bring upon themselves!
Manton’s takes on “AI” specifically and big companies in general exhibits a level of poor judgement that makes me really worried as an active user of micro.blog
And his poor judgement has been quite consistent for a while now. It just seems to be more on the surface lately
Like this earlier post on the Scarlett Johansson thing where he talks about ‘an “OpenAI can’t trusted” narrative’, which would imply that the only real problem is how OpenAI’s behaviours are framed, that the actual behaviours themselves are fine.
At least, the potential for abuse should be obvious to somebody who runs a social network that puts privacy and control over your own data front and centre. That he doesn’t see the potential issues makes you wonder how careful he’s being about privacy issues on micro.blog