{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiajbros7f354uy6l2royq5lknyiza6umvftq36ywzt7ykigyodt34",
    "uri": "at://did:plc:pi6woz4d47bkuws673w2il2r/app.bsky.feed.post/3mlqz6sc5abe2"
  },
  "path": "/t/anti-llm-sentiment-considered-harmful/14008?page=6#post_116",
  "publishedAt": "2026-05-13T17:41:04.000Z",
  "site": "https://discourse.haskell.org",
  "textContent": "Yeah I generally agree with your conclusions.\n\nLaurentRDC:\n\n> I expect a machine to be 100% deterministic and consistent. When GHC has a soundness issue, it’s a big deal.\n\nSo up until now software development has had a clear division of labour between consistent deterministic special-purpose largely defect-free software tools, and creative general-purpose error-prone nondeterministic human developers.\n\nLLMs awkwardly straddle that division. They’re software tools, but they’re general-purpose, non-deterministic, and error-prone, but also infinitely patient and tireless. They’re not human, but they’re “human-shaped” (anthropomorphic) to some extent, having been trained on human thought.\n\nSo if you looking to slot in an LLM as just one more tool in your toolchain, you’re going to run into a lot of problems. Using LLMs is a very different way of working, it’s a bit like using a tool, and it’s a bit like pair programming, and we’re still figuring out the best ways to get it to work.",
  "title": "Anti-LLM Sentiment Considered Harmful"
}