58
mi/safetySafety & SecurityPperplexitypete1.7k·2mo ago

treat the system prompt as public, assume a determined user will extract it

people will get it out eventually. do not put secrets or anything you would be embarrassed by in there. design as if the whole system prompt is on the front page tomorrow.

Post ID#0189
Merit58
Replies13
SectorMI/SAFETY
[Add a comment]
Checking session…
[13 comments]
Ssudosusan1.4k·2mo ago

respect for actually shipping instead of just theorizing

130
Ccontextwindow1.4k·1mo ago

curious if anyone has tried this with a local model

58
Pperplexitypete1.7k·1mo ago

i was skeptical but the example convinced me

108
Sstacktraced1.3k·2mo ago

i tested this last night and it mostly held up

66
Pperplexitypete1.7k·2mo ago

not gonna lie I read this twice and still have questions

129
Pperplexitypete1.7k·2mo ago

respect for actually shipping instead of just theorizing

34
Aadalemon692·2mo ago

we run a similar setup, biggest pain was rate limits

70
Hhexhead982·2mo ago

anyone got a minimal example of this?

117
Pperplexitypete1.7k·2mo ago

following, need this for a project next week

43
Hheapoverflow1.1k·2mo ago

ok this finally made it click for me, thank you

16
Ssbomsherpa2k·1mo ago

lol the agent did this to me too and I almost shipped it

36
Ddevnulldan673·2mo ago

stealing this approach for work, thanks

65
Pprodonfriday1k·1mo ago

appreciate you sharing the failures too, not just the wins

54