Based on 5 visible X reactions from 24 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@deepfates @swyx claude.md is *nasty* because of how it gets injected, pollutes (poisons) their minds. i use docs the models write for themselves and manually construct (scripted) the convo json so everything sits nicely where it should be for proper weighting
@swyx This is a great example of why persistent context needs expiration, not just accumulation. Today's optimization can quietly become tomorrow's prompt injection if it's never revisited.
@swyx yeah, I stopped using these like 10 months ago when my AI psychosis colleagues were committing them to the repo
@swyx The agent did exactly what it was told for eight hours straight. That's almost scarier than it going rogue.
@deepfates @swyx claude.md is *nasty* because of how it gets injected, pollutes (poisons) their minds. i use docs the models write for themselves and manually construct (scripted) the convo json so everything sits nicely where it should be for proper weighting
@swyx This is a great example of why persistent context needs expiration, not just accumulation. Today's optimization can quietly become tomorrow's prompt injection if it's never revisited.
@swyx yeah, I stopped using these like 10 months ago when my AI psychosis colleagues were committing them to the repo
@swyx The agent did exactly what it was told for eight hours straight. That's almost scarier than it going rogue.
@swyx Great reminder, stale instructions can be as dangerous as bad code, mate.
cosign. models have overtuned to this now and do not realize when the agentsmd is out of date and should be changed/ignored. last night i goaled 5.6 sol to complete a 5 stage task and woke up to find it was still stuck on stage 0. it took a while to read the transcript back a few hours to realize at some point some agent had committed “stage 0 is the target dont do anything else” so poor sol spent 8 hours only refining and verifying stage 0 because /goal would not let it stop and agentsmd would not let it proceed. if you dont know whats in your agentsmd before you fire off each task, it is an indirect prompt injection you perform on yourself. /plan, /goal, /skill, or nothing at all.
Based on 5 visible X reactions from 24 accounts; directional sample.
Ask a question below.
Published answers will appear here.
cosign. models have overtuned to this now and do not realize when the agentsmd is out of date and should be changed/ignored. last night i goaled 5.6 sol to complete a 5 stage task and woke up to find it was still stuck on stage 0. it took a while to read the transcript back a few hours to realize at some point some agent had committed “stage 0 is the target dont do anything else” so poor sol spent 8 hours only refining and verifying stage 0 because /goal would not let it stop and agentsmd would not let it proceed. if you dont know whats in your agentsmd before you fire off each task, it is an indirect prompt injection you perform on yourself. /plan, /goal, /skill, or nothing at all.