perching_aix 3 days ago

They're saying that it successfully filtered out the bit where the author told people to overdose by 40000x. I guess that's the value.

Reply View 3 replies
  • phoronixrly 3 days ago

    There would be value if it pointed out the mistake instead of hallucinating a correction.

    Reply View 2 replies
    • naasking 3 days ago

      IU was correctly used everywhere else in the article except that one place with the mistak, so the LLM didn't hallucinate a correction, it correctly summarized what the bulk of the article actually said.

      Reply View 0 replies
    • pulvinar 3 days ago

      GPT5.2 does catch it and warns to not trust anything else in the post, saying no competent person would confuse these units.

      I wonder if even the simplest LLM would make this particular mistake.

      Reply View 0 replies