behdad (@[email protected])

HarfBuzz 13.0.0 released with new experimental features: `hb-vector` for vector output of glyph outlines to SVG and `hb-raster` for rasterizing glyphs to A8 / BGRA32 images, among other features. I would like to welcome our new contributors: Claude & Codex, without whom these new features would not have been possible. https://github.com/harfbuzz/harfbuzz/releases/tag/13.0.0

typo.social
@davidgerard Why do so many presumably smart people fall for the LLMs? It’s disappointing to say the least.
@schrotthaufen @davidgerard The AI fad seems to infect mostly those who have either a dangerously high or dangerously low amount of self esteem 🤷‍♀️
@schrotthaufen @davidgerard Because they're designed to exploit what are basically security vulnerabilities in the human brain, specifically ones that another human can't easily exploit, which we have no defense against because the ones explained by humans are the only ones we've had to build an immunity to! See slot machine addiction as another example of machines designed to exploit your brain's security holes.
@jackemled @schrotthaufen @davidgerard
They're the first actual memetic weapons ever invented.
@kirtai @schrotthaufen @davidgerard No, propaganda has existed for centuries. This might be the second ever though.
@jackemled @schrotthaufen @davidgerard
True, but I was thinking more of actual objects than concepts.

@jackemled @schrotthaufen @davidgerard How did they might be giants put it?

A woman came up to me and said
"I'd like to poison your mind
With wrong ideas that appeal to you
(Al)though I am not unkind."

@schrotthaufen @davidgerard I've been avoiding these tools for lots of reasons, and I'd understand if nobody else on this thread knows because they've also been avoiding them, but just in case not: do software development-oriented generative-"AI" tools like Claude Code or Codex actually have the extreme people-pleasing suck-up mode that stock ChatGPT has? I'm trying to figure out what class of cognitohazard I'm dealing with when I see so many people saying things about how amazing coding generative-"AI" is.

@schrotthaufen @davidgerard

And I think I’ve got my answer. Check out this AI-slop-laden pull request for changes to vim. Yes, vim, the old-school text editor that’s been around for decades and historically caters to people who don’t like things like generative “AI”:

https://github.com/vim/vim/pull/19413

The submitter of the PR is responding to comments with responses that are obvious AI sludge, probably just cut and pasted from the chat window, and the beginning of every one of them is effusive praise of the sort only used-car salesmen, overt con artists, and LLMs can pull off multiple times in a row.

Assuming the author of the PR was using the same application for the comments as they did for the original PR, then, yes, coding-oriented LLMs are very much indeed doing the same sycophantic shit as ChatGPT 4o, just maybe dialed back a teensy bit.

#AIslop

Add setrepeat() and getrepeat() functions for dot command control by Shougo · Pull Request #19413 · vim/vim

Summary Add setrepeat() and getrepeat() functions to allow scripts to programmatically control the dot (.) repeat command. This enables plugins to: Save and restore the repeat command Make custom ...

GitHub

@dpnash @schrotthaufen @davidgerard

And the PR was reviewed by Copilot *bangs face on floor*
Vim is my daily driver for fucksake. Please don't!

I'm going to look at my staring wall and wait for oblivion

@bloognoo @dpnash @schrotthaufen @davidgerard

just thought it was worth pointing out this reply for their perspective on it

https://github.com/vim/vim/pull/19413#issuecomment-4013824271

Add setrepeat() and getrepeat() functions for dot command control by Shougo · Pull Request #19413 · vim/vim

Summary Add setrepeat() and getrepeat() functions to allow scripts to programmatically control the dot (.) repeat command. This enables plugins to: Save and restore the repeat command Make custom ...

GitHub

@gloriouscow @dpnash @schrotthaufen @davidgerard

That's far more reassuring than typical responses, I can live with that.

!!!EDIT!!!

No I fucking can't. 'CoPilot doesn't like this'?!

Trust but verify is a floor not a ceiling.

There is a line between using AI to triage incoming PR then investigating what looks not to be crap that is just tolerable (but not sustainable). Handing off the responsibility for a decision as if CoPilot was a team member or some kind of oracle* is NOT.