4 ms·

I believe it is extremely important to disclose that the ‘responses leaks’ you obtained did not originate from LLM models themselves, but rather through other i

by JyB 1y ago

I believe it is extremely important to disclose that the ‘responses leaks’ you obtained did not originate from LLM models themselves, but rather through other insecure systems / in a more conventional manner. Just to avoid yet another case of hallucinations outputs getting misinterpreted.

requilence 1y ago
Right, thank you for the suggestion. Just added a paragraph to the original blog post.
- tabletcorry 1y ago
  Your added paragraph appears to suggest the opposite, that this was an LLM response. Was the "leaked data" a response from an LLM directly?
  - JyB 1y ago
    Yes apparently which makes this report pretty flimsy.
    - tptacek 1y ago
      Upthread, OpenAI's security team confirms it's a false report; it's a variant of the empty-prompt hallucination.
      - JyB 1y ago
        Incredible that so many people still don't understand what an LLM is. Especially ones that you would expect to grasp it.