GitHub claimed over 50% of code was now being done with LLM tools. Further information shows that is not exactly the full story.
The Vervint article in the introduction there says: “GitHub’s own data shows that while Copilot might suggest code for 35-40% of what a developer types, only about 26% of those suggestions get accepted. Furthermore, we have far less data on how much accepted code actually survives code review, testing, and makes it to production unchanged.” It continues saying: “Context matters enormously. For example: A junior developer working on backend CRUD operations of an app might accept 40% of AI suggestions, while a senior engineer architecting complex distributed Agentic AI systems might accept less than 10%. The type of project, programming language, and company culture all dramatically impact these numbers.”
A comment reply to a news.ycombinator.com submission I made for a Techrights article makes a point that I’ve made before in discussions. People with the means to actually do so may try using the LLM to write a new GNU/Linux like OS without the GPL. Whether it will work is something we should have doubts on, as it likely won’t even reach the sloppy slopped bootloader when tested.
We know why IBM and RedHat, Canonical, Microsoft, Oracle, Google, and the rest of big tech and big FOSS are doubling down on LLM slop: they want to find people to give free money as a shareholder. But, like Roy wrote: “It cheats the shareholders, but for how much longer?”
What I can say now, at least, is that it sure would be horrible to have to have no way to easily get away from constant slop articles, slop posts, slop code, and blatant slop promotion. Consider news/information sources and software sources that do more than just LLM slop. What a bad time to be in (or going into) a tech industry right now.