Lightnews — Scholar-powered news

Joe Fabisevich

@mergesort.me

3.9K followers 71 following 4.3K posts

“Apps genius” - Colleen. I used to work on making Twitter a bit healthier, then that all went to hell. Now I make puns and indie apps @plinky.app. Born and raised New Yorker, trying to do a little good and be the friend you made along the way. 🍕🐱⚾️

Posts Media Videos Starter Packs

Pinned

Joe Fabisevich @mergesort.me · Nov 9

Hi, I’m Joe! I used to work @ Twitter on all the hard things like misinformation, disinformation, harassment, and civic integrity. I also spent a ton of effort on little improvements to make people’s lives better. These days I’m an indie developer making it easy to save links for later, with Plinky.

Plinky

Plinky helps you easily save links from any device, so you can enjoy them later. Save on your iPhone, iPad, Chrome, Safari, Firefox, Zapier, and so much more.

plinky.app

5 12 110

Joe Fabisevich @mergesort.me · 1h

Now that we have a name 50% of the work is done. I’m a little mad that I have absolutely no time right now, otherwise I’d quickly whip something up with Codex.

Joe Fabisevich @mergesort.me · 1h

Lorem atsum

1 9

Joe Fabisevich @mergesort.me · 23h

I'm kinda done debating because you're making it clear you didn't even read the research you cite. (I actually have read the papers and you can't extrapolate from 16 developers in a randomized trial, 8 of whom had never used Cursor — a few of didn't even know the features Cursor offers.)

Ruben Bloom (Ruby) on X: "I was one of the developers in the @METR_Evals study. Thoughts: 1. This is much less true of my participation in the study where I was more conceintious, but I feel like historically a lot of my AI speed-up gains were eaten by the fact that while a prompt was running, I'd look" / X

I was one of the developers in the @METR_Evals study. Thoughts: 1. This is much less true of my participation in the study where I was more conceintious, but I feel like historically a lot of my AI speed-up gains were eaten by the fact that while a prompt was running, I'd look

x.com

Joe Fabisevich @mergesort.me · 23h

I’m well aware of this subject matter but what you’re describing is a different problem. For example workslop is a matter of laziness and incentives, mediocre results from people who don’t use AI in a novel way but instead regurgitate whatever it says verbatim. That’s not who I’m talking about.

Joe Fabisevich @mergesort.me · 23h

Current models simply do not hallucinate 30% of the time. If you believe that than odds are you haven’t tried any model that was released in the last six months, because the amount of progress they’ve made on the hallucination front is wildly good, especially for software development subjects.

2 1

Joe Fabisevich @mergesort.me · 23h

Nothing was stopping them, but people learn differently. Have you ever met a developer who learns better from video than reading? Or by doing than sitting in a lecture? There are a lot of people who will learn better from AI (and many worse or far worse), but you can’t disqualify them if they learn.