2026-02-20 20:12 UTC
@fasterandworse This is not my strongest objection to LLMs and "generative AI", but it's the one that it puzzles me the most that people, like Doctorow, and like professional software developers, do not see: tools are supposed to be predictable and reliable. The discipline of engineering is supposed to be founded on this principle. But this is a class of software that is, by design, unpredictable and unreliable. (Even random number generators generate numbers in a predictable range and frequency.) They are not tools, and should not be used.
My stronger objections have to do with the fascist ideology built into their design, and the colossal ecological and economic costs of creating them.
But even in an abstract, ideal case where those objections were overcome, the fundamental conceptual flaw remains. They aren't proper tools, and anyone who understands the ethical precepts of engineering should reject their use.
Replies (2)
-
@oborosaur@hachyderm.io 2026-02-20 21:46
@foolishowl @fasterandworse it does nothing to address the ideological, ecological, and economic concerns, but I found the Oxide and Friends discussion about using LLMs to support engineering rigour contained a number of thoughtful examples that do not require perfect precision and accuracy: https://oxide-and-friends.transistor.fm/episodes/engineering-rigor-in-the-llm-age
-
@davidculley@hachyderm.io 2026-02-21 01:07
@foolishowl @fasterandworse Yes. I don't see the point in using a LLM to check my text for typo's when there are lots of deterministic tools for exactly this purpose, like LanguageTool, Vale with Openly (open alternative to Grammarly), Hemingway, and many others like them. https://languagetool.org https://vale.sh