What question did this study set out to answer?

To investigate the effectiveness of closed-loop binary conditioning in transferring operator states across different LLM architectures.

February 26, 2026Open Access

Closed-Loop Binary Conditioning Produces Transferable Operator State Across LLM Substrates

Key Points

To investigate the effectiveness of closed-loop binary conditioning in transferring operator states across different LLM architectures.
Developed a real-time closed-loop protocol for LLM generation based on binary accept/reject judgments.
Compiled a persistent ledger to update the state through gradient descent on binary labels.
Conducted independent runs across four distinct LLM architectures.
19 out of 20 independent runs successfully learned the same preference directions.
The compact operator state converges from a cold start across different LLMs.
The protocol facilitates real-time adaptation without needing model weight updates.

Abstract

We introduce a real-time closed-loop protocol in which each LLM generation isconditioned on a persistent ledger of binary accept/reject judgments accumulatedwithin a session. The ledger state is updated by gradient descent on the binarylabels, yielding a compact operator state that converges from cold start andtransfers across four architecturally distinct LLMs — 19 of 20 independent runslearn the same preference directions. These findings suggest that persistentin-session conditioning can serve as a complementary interface-level mechanismto offline preference optimization, enabling real-time adaptation of generationdynamics without model weight updates

Closed-Loop Binary Conditioning Produces Transferable Operator State Across LLM Substrates

Key Points

Abstract

Cite This Study