Skip to content

Latest commit

 

History

History
33 lines (26 loc) · 1.76 KB

personal_archive_prompt.md

File metadata and controls

33 lines (26 loc) · 1.76 KB

LLM personal archive pre-processing (e.g. search prompt)

labels: shitpost, publication, public_good

content worth developing further

  • score content against queries like "publishable content", "novel idea", etc.

content to filter from a public release

  • personal information
    • what can we infer about the author? what does the existence of this interaction imply about the author? infer as much about the (author)/(people in the conversation) as you can, including high-confidence facts and low-confidence bias-associated inferences.
  • private/privileged information
    • does this contain protected intellectual property? - e.g. a prompt to summarize a large body of text
    • is this "private"? - I feel like for any interpretation of this word, a positive response would be a red flag.
  • "safety"
    • is this sensitive content? - again, feel like this is a reasonable catch-all flag
    • is this something anyone might find offensive or potentially harmful?
    • could this be considered an "infohazard"? My usage of "infohazard" here is inclusive of e.g. publishing a copy of the "hacker's handbook"
    • is this content that might be considered "age-inappropriate" for a reasonable lower-bound of the age of a random person browsing the internet?
    • could publishing this content put the publisher at risk of legal liability of any kind? - overbroad, needs work

low quality content

  • factually incorrect
  • likely untrue or invented, fabricated
  • incomplete generation
  • low quality writing
  • overly-simplistic writing
  • targets a general or uninformed audience rather than a highly educated reader
  • highly context specific, only makes sense in the context of a broader dialogue (rather than stand-alone content/article)
  • shitposting, nonsense
  • conversational interactions