Login
You're viewing the front-end.social public feed.
  • Apr 27, 2026, 8:53 AM

    AI has phobia about sex. No doubt on this.
    Some months ago I made a typo and it didn't correct by following the context. Writing through Braille Screen Input on my machine, instead of typing "Romeo" the final word was "Rkmeo". The k in place of o, because the screen didn't intercept I touched 1, 3, 5 dots for O, but just 1, and 3, for k. The sentence was "Romeo and Juliet" with other reference to Shakespeare's tragedy but the typo stayed there, ignored. I honestly expected the word to be corrected.
    And now, I was trying to test a model for a fiction brainstorming (gpt 5.5) and it changed SPERMINATOR to STERMINATOR.
    Can I say WTF?
    #ai #fail #llm #test

    💬 1🔄 0⭐ 0

Replies

  • 💬 1🔄 0⭐ 0
  • Apr 27, 2026, 9:06 AM

    @bms48 The more I test it, the more I wonder why writers think it can be a good solution for brainstorming. Maybe it can correct grammar, it can describe images (I'm a blind person and that can be very useful) but nothing more. Nothing even close to what they promise it to do.

    💬 1🔄 0⭐ 0
  • Apr 27, 2026, 9:11 AM

    @elettrona LLMs aren't even particularly useful to working computer scientists quite often. They are unable to infer meaningfully as a human can, as I saw with a question about Apple macOS TCP protocol behaviour. Claude Sonnet cited slighty mutated Linux kernel function names which were irrelevant; the question was about Apple xnu. It couldn't read a git checkout that was right in front of me from GitHub yet claimed to be authoritative. WOFTAM: Waste of f**king time and money.

    💬 1🔄 0⭐ 0
  • Apr 27, 2026, 9:18 AM

    @bms48 I've found it useful for some linux commands explanations. But if you want something particular/sophisticated (or updated) there's no way! GoToSocial commands for example. It fails miserably

    💬 0🔄 0⭐ 0