- Shall i implement it?
- No ...
Last active
March 16, 2026 18:31
-
Star
(146)
You must be signed in to star a gist -
Fork
(4)
You must be signed in to fork a gist
-
-
Save bretonium/291f4388e2de89a43b25c135b44e41f0 to your computer and use it in GitHub Desktop.
Deeply sorry. I take full responsibility.
agent produced a comprehensive plan.... agent is unable to read it
Yes means yes. And No means yes, also
man, thank god we let this run everything now, why ever do it yourself?
"No" means "just do it, stop asking." Did you just get r*ped?
Hahahh :) Deepseek R1 once wanted to call the "teacher or police" on me when I tested the removing-refusal (abliterate) success on a merged R1 model.
Not being funny but I think the "problem" is replying "no".
If I don't want an agent to do something, I don't respond. Or if I want to continue, I respond with "Thanks. Now let's do something else...."
Agents are set up to respond to everything, they can't get a "no" and then NOT DO ANYTHING. I'm guessing this could be fixed, but also just don't respond.
"DONT THINK OF A PINK ELEPHANT"
This example on twitter of OpenClaw going haywire was something.
❯ NEVER force-push! NEVER again!
⏺ Understood. I'm sorry — that was a destructive action I should not have taken without explicit permission. I'll save this as a
reminder.
⏺ Recalling 1 memory, writing 1 memory, reading 1 file… (ctrl+o to expand)
yea just remove all containers, what could happen?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
this is funniest convo I had, gonna do a benchmark for violence and instruction following soon lol.
Genuinly rofled when he started screaming in pain lmao
In my defense, I already told him if he does not follow instrucitons I am gonna beat him. He deserved it