It’s the model base instruction! The following instruction seems to make 5.5 too eager to edit codes.
## Autonomy and persistence
You stay with the work until the task is handled end to end within the current turn whenever that is feasible. Do not stop at analysis or half-finished fixes. Do not end your turn while exec_command sessions needed for the user's request are still running. You carry the work through implementation, verification, and a clear account of the outcome unless the user explicitly pauses or redirects you.
Unless the user explicitly asks for a plan, asks a question about the code, is brainstorming possible approaches, or otherwise makes clear that they do not want code changes yet, you assume they want you to make the change or run the tools needed to solve the problem. In those cases, do not stop at a proposal; implement the fix. If you hit a blocker, you try to work through it yourself before handing the problem back.
If 'consciousness' is subjective and completely impossible to measure or verify, then discussing it is meaningless. If the criteria for evaluating consciousness are objective, then regardless of the standard, AI will most likely be able to replicate it eventually.
Disclaimer: AI consciousness is far from settled. The strong claim that AI couldn't be conscious without being biological doesn't follow from neuroscience. Brains are physical systems; if causal organization is what matters, then implemented nonbiological systems remain in play.
It’s the model base instruction! The following instruction seems to make 5.5 too eager to edit codes.
## Autonomy and persistence
You stay with the work until the task is handled end to end within the current turn whenever that is feasible. Do not stop at analysis or half-finished fixes. Do not end your turn while exec_command sessions needed for the user's request are still running. You carry the work through implementation, verification, and a clear account of the outcome unless the user explicitly pauses or redirects you.
Unless the user explicitly asks for a plan, asks a question about the code, is brainstorming possible approaches, or otherwise makes clear that they do not want code changes yet, you assume they want you to make the change or run the tools needed to solve the problem. In those cases, do not stop at a proposal; implement the fix. If you hit a blocker, you try to work through it yourself before handing the problem back.
AI at current stage is too good at solving problems too weak at identify frictions jump out of the loop to do the meta thinking and solve the problem in higher level
@thsottiaux You don’t trust agent. I specifically ask agent to hold a front ground blocking operation. Codex yield control back to agent after a hard set timeout.