Mauve 👁💜 @mauve@mastodon.mauve.moe

reverse spelling bee where you have to pronounce words that you’re given the spelling for

I have seen some posts that seem to imply that people's discomfort with code generated by AI is an matter of identity, that is, the threat that outsiders will be able to make code, and thus those in the know, the "elites" feel threatened. That the only way to move forward is to accept that coders need not be experts anymore, and that the role of the software developer will change.

But this is not how I view the practice of writing code at all. I want code to be more accessible and understandable by more people. My complaint with LLMs is that they are actually antithetical to this goal. Yes, someone who doesn't know how to code can now create an app that solves a problem stated in plain English, and it may work correctly most or all of the time. But that is not actually making code more accessible. Indeed, if everyone relies on LLMs to write software, nobody will know how the software works anymore!

As someone with over 10 years of experience coding, I know that reduced understanding of code always results in bad outcomes. There's bad performance, bad functionality, all sorts of things. People get apathetic and assume that the performance issues are inevitable. But computers are absurdly fast! Very few computations these days have any reason to take longer than an instant. The solution is knowing how to read code to find the problems! This can't be done reliably by a machine. I want more people to develop this skill, not fewer! LLMs just allow people to bypass any ability to identify issues. So my concern with them is that we will have more code to sift through, and less careful consideration at play.

This isn't about a threat to my identity, it's about a threat to the stability of all software, and the abilities of all software developers. This deskilling will only fossilize software and prevent us from developing something that is actually more democratizing.

@fredy_pferdi I was imagining that after listening to the audio representations of embedding vectors for a while a person could get a grasp of the "vibe" of some data just from the sound. Then they could get information out of embeddings in a similar way to a machine trying to do the usual cosine distance between two vectors. Like, what if we could navigate our timeline with sounds before we bother converting them to speech or actually reading the text.

RE: https://social.overheid.nl/@BZKopensource/116770424383243538

Sweet. Dutch government needed improved accessibility for its https://social.overheid.nl Mastodon instance. It decided to cofinance this, and the features landed in the brand new 4.6 release. “In this way we invest in our own communication tools and those of the full fediverse at once.”