Do Language Models Know Their Own Constraints?
TL;DR
Comparing supervised fine tuning and Group Relative Policy Optimization against an untrained baseline on a four tier Constraint Awareness Benchmark shows the reward based signal is the more destructive of the two: a reward that penalizes banned ingredient tokens regardless of framing learns a context independent suppression rather than a self directed constraint.
Abstract
We ask whether behavioral constraints acquired through post training remain explicitly reportable. Using constrained recipe generation as a testbed, five banned ingredients enforced via LoRA fine tuning of Llama 3.1 8B Instruct we compare supervised fine tuning (SFT) and Group Relative Policy Optimization (GRPO) against an untrained baseline on a four tier Constraint Awareness Benchmark. Averaged over three seeds, both methods raise behavioral compliance from 4% to about 90% while reducing explicit constraint reporting below the untrained model (0.48/5 to 0.16/5 for SFT, 0.07/5 for GRPO) and eroding retained third person knowledge (93% to 36% for SFT, 14% for GRPO; p less than 0.01 between methods). Contrary to our initial hypothesis, the reward based signal is the more destructive of the two: a reward that penalizes banned ingredient tokens regardless of framing learns a context independent suppression rather than a self directed constraint. A context conditioned reward designed to teach the self to other distinction fails, collapsing toward inclusion in both framings. Probing prompt time hidden states recovers per ingredient avoidance at 83.8% (layer 24 MLP), but only 6.4 points above a per ingredient base rate predictor (77.4%), and the model's own verbal self report is more accurate still (87.8%). A positive control adding explicit self description examples does not restore reporting. The failure is therefore specific to enumerating constraints on request, not a general loss of access to them.
