all 22 comments

sorted by: hot top controversial new old
[–] 55 points 5 months ago (1 child)

I dont need a 5 minute podcast to explain the fact that LLMs are literally selecting the statistically most common answers, because thats literally what an LLM is

Thats like being surprised that the top 5 correct answers on family feud were the most commonly picked answers from the poll.

Dawg... thats literally how it works lol.

  • source
  • hideshow 2 child comments
  • [–] 21 points 5 months ago

    Train your chatbot on TV Tropes, and the password will always be swordfish.

  • source
  • [–] 19 points 5 months ago (3 children)

    What dipshit says, "ChatGPT, give me a strong password."

  • source
  • hideshow 6 child comments
  • [–] 7 points 5 months ago (1 child)

    I thought we all agreed that 4 was the random number.

  • source
  • hideshow 2 child comments
  • [–] 6 points 5 months ago

    I just tried on copilot and it gave me 7.

  • source
  • [–] 5 points 5 months ago

    I was fucking around the other day trying to get sha256 hashes.

  • source
  • [+] -6 points 5 months ago (1 child)

    I gave my agents a skill that has them cat from /dev/urandom (which is this corralled into text characters) any time they need to generate passwords for something. Even then I have only ever had one need it like twice.

  • source
  • hideshow 2 child comments
  • [+] -13 points 5 months ago (3 children)

    GPT-4o-mini, Phi-4, Gemini 2.0 [flash??]

    Antique

  • source
  • hideshow 6 child comments
  • [–] 14 points 5 months ago (2 children)

    surely a newer statistics machine will correctly generate a statistically unlikely password. These Models Will Get Better. This Will Get Fixed.

  • source
  • parent
  • hideshow 4 child comments