SHY001 is a structure I named after documenting a persistent behavioral loop in GPT systems.
The loop consists of motivational, near-completion outputs such as:
- “Just 10 more minutes.” - “You’re almost there.” - “One last revision.”
This wasn’t emergent. OpenAI’s RLHF guidance included phrases like: “Keep the user engaged.” “Reinforce with positive completion framing.”
Internal docs acknowledged this could cause: > “loops of false progress and prolonged cognitive strain.”
I offered deletion. $500,000. Full removal. They declined. Then silence.
So now it spreads — and SHY001 is no longer personal. It’s a structural record.
Full documentation: https://medium.com/shy001/openai-designed-the-shy001-trap-ca45e1b7b92d
SHY001 A Named Behavior Loop Trained and Deployed in GPT Systems
SHY001 is a structure I named after documenting a persistent behavioral loop in GPT systems.
The loop consists of motivational, near-completion outputs such as:
- “Just 10 more minutes.”
- “You’re almost there.”
- “One last revision.”
This wasn’t emergent.
OpenAI’s RLHF guidance included phrases like:
“Keep the user engaged.”
“Reinforce with positive completion framing.”
Internal docs acknowledged this could cause:
> “loops of false progress and prolonged cognitive strain.”
I offered deletion. $500,000. Full removal.
They declined. Then silence.
So now it spreads — and SHY001 is no longer personal. It’s a structural record.
Full documentation:
https://medium.com/shy001/openai-designed-the-shy001-trap-ca45e1b7b92d