Elektrine lite

← Feed

@lina@neuromatch.social

2026-08-31 01:55 UTC

We get the typical "slop can be good sometimes" for a vaguely defined set of use cases... "AI tools can sometimes be beneficial, particularly when it comes to finding patterns among a large data set (entire code base), performing tedious repetitive changes or large refactoring/re-organization." and also this gem... "AI tools are treated the same as traditional tooling like sed, awk or coccinelle." which is a reversal of how they treated LLMs in their previous "No Large Language Models (LLMs) or similar AI tools" statement: "All contributions to this project are expected to be done by human beings or through standard predictable tooling [...] LLMs and similar predictive tools have the annoying tendency of producing large amount of low quality code with subtle issues which end up taking the maintainers more time to debug than it would have taken to write the code by hand in the first place." Maybe they conflated "predictable tooling" with "predictive tools" (already a slightly sinister framing). And I guess subtle code issues are no longer a concern, since they're only suggesting that LLMs are useful for "boring grunt work" like (checks notes) "large refactoring" 🤦‍♀️ well, I'm sure the subtle issues will never slip past review. 2/🤡

Replies (1)

  • @lina@neuromatch.social 2026-08-31 01:55

    They state that LLMs "must abide by our instructions (AGENTS.md)". Of course, we have uncountable examples of "guardrails" being trivially circumvented, both intentionally and unexpectedly. As @olivia@scholar.social, Marcela Suarez, and @Iris@scholar.social say in Towards Critical Artificial Intelligence Literacies: "Guardrails may seem sensible, giving the impression of responsibility, but critical computational thinking through formal analyses reveal they are unimplementable." https://doi.org/10.5281/zenodo.17786243 And as @davidgerard@circumstances.run recently reported, it seems that AGENTS.md files don't do much beyond giving a pseudo-IaC aesthetic to your "make no mistakes" prompts. https://pivot-to-ai.com/2026/08/27/your-agents-md-file-doesnt-actually-do-anything/ Hilariously, their AGENTS.md includes a legal section with the impossible prompts... "All contributions to this repository must be compatible with the Apache 2.0 license [...] contributions cannot include code licensed under the terms of the GPL, AGPL or LGPL licenses." Other sections include a few formatting and testing/validations prompts. 3/💩

    Open ##4582822