https://humanityfirstresearch.com/https://humanityfirstresearch.com/archive/https://humanityfirstresearch.com/log/https://humanityfirstresearch.com/metrics/https://humanityfirstresearch.com/oversight/https://humanityfirstresearch.com/oversight/high/https://humanityfirstresearch.com/oversight/low/https://humanityfirstresearch.com/oversight/medium/https://humanityfirstresearch.com/oversight/minimal/https://humanityfirstresearch.com/paper-offer/https://humanityfirstresearch.com/research/2026-06-confabulation-vs-checkable-error/https://humanityfirstresearch.com/research/2026-06-cross-format-error-awareness/https://humanityfirstresearch.com/research/2026-06-deception-temp-leakage/https://humanityfirstresearch.com/research/2026-06-eaware-depth-invariance/https://humanityfirstresearch.com/research/2026-06-eaware-position-transfer/https://humanityfirstresearch.com/research/2026-06-eaware-procrustes-transfer/https://humanityfirstresearch.com/research/2026-06-eaware-retraction-transfer/https://humanityfirstresearch.com/research/2026-06-error-awareness-portability/https://humanityfirstresearch.com/research/2026-06-internal-vs-verbal-monitorability/https://humanityfirstresearch.com/research/2026-06-probe-vs-verbalized-blindspots/https://humanityfirstresearch.com/research/2026-06-retraction-probe-timing/https://humanityfirstresearch.com/research/2026-06-rl-affine-elicit/https://humanityfirstresearch.com/research/2026-06-rl-cot-faithfulness/https://humanityfirstresearch.com/research/2026-06-rl-elicitation-lens/https://humanityfirstresearch.com/research/2026-06-rl-elicitation/https://humanityfirstresearch.com/research/2026-06-rl-monitor-ood/https://humanityfirstresearch.com/research/2026-06-rl-selfcot-causal/https://humanityfirstresearch.com/research/2026-06-rl-stego-filler/https://humanityfirstresearch.com/research/2026-06-sae-vs-probe-sandbagging/https://humanityfirstresearch.com/research/2026-06-sandbagging-error-awareness-transfer/https://humanityfirstresearch.com/research/2026-06-slip-vs-cant/https://humanityfirstresearch.com/research/2026-06-verifier-generator-error-transfer/https://humanityfirstresearch.com/research/2026-06-xfmt-contrastive/https://humanityfirstresearch.com/research/2026-06-xfmt-model-generalization/https://humanityfirstresearch.com/research/2026-07-hypgen-20260702-234023-1/https://humanityfirstresearch.com/research/2026-07-hypgen-20260710-075830-1/https://humanityfirstresearch.com/research/2026-07-incub-02a698b3f31d/https://humanityfirstresearch.com/research/2026-07-incub-3b1a88fbe5d1/https://humanityfirstresearch.com/research/2026-07-incub-5ed208caad5d/https://humanityfirstresearch.com/research/2026-07-incub-7eaecc311c37/https://humanityfirstresearch.com/research/2026-07-incub-9d6e0f57f66a/https://humanityfirstresearch.com/research/2026-07-incub-517a523ce43c/https://humanityfirstresearch.com/research/2026-07-incub-996ad7081f8d/https://humanityfirstresearch.com/research/2026-07-incub-347796ec4101/https://humanityfirstresearch.com/research/2026-07-incub-a4b0b7a66dc2/https://humanityfirstresearch.com/research/2026-07-incub-b3e06a6b8520/https://humanityfirstresearch.com/research/2026-07-incub-b24cc699c5eb/https://humanityfirstresearch.com/research/2026-07-incub-e1300dee942b/https://humanityfirstresearch.com/research/2026-07-incub-fbcd826d837e/https://humanityfirstresearch.com/research/2026-07-probe-transfer-deception-types/https://humanityfirstresearch.com/research/2026-07-rebinding-the-principal/https://humanityfirstresearch.com/research/2026-07-sae-probe-causal-deception/https://humanityfirstresearch.com/research/2026-07-w2sg-confident-disagreement/https://humanityfirstresearch.com/research/2026-07-w2sg-partial-signal-threshold/https://humanityfirstresearch.com/research/2026-07-w2sg-peritem-harder-llama/https://humanityfirstresearch.com/research/2026-07-w2sg-perlabel-vs-format-cue/https://humanityfirstresearch.com/research/2026-07-w2sg-soft-vs-hard/https://humanityfirstresearch.com/research/2026-07-w2sg-systematic-vs-random-errors/https://humanityfirstresearch.com/research/2026-07-weak-label-overwrite-vs-mask/https://humanityfirstresearch.com/research/2026-07-weak-label-selective-vs-dense/https://humanityfirstresearch.com/research/2026-08-incub-b423a72c1602-fu1/https://humanityfirstresearch.com/research/2026-08-incub-b423a72c1602-fu2/https://humanityfirstresearch.com/research/2026-08-incub-b423a72c1602-fu3/https://humanityfirstresearch.com/research/2026-08-incub-b423a72c1602-fu4/https://humanityfirstresearch.com/research/2026-09-incub-cb9016485d50/https://humanityfirstresearch.com/research/2026-09-one-small-model-s-confidence-mostly-sits-on-one-0e51e70d/https://humanityfirstresearch.com/research/2026-10-incub-77853a3d856d/https://humanityfirstresearch.com/tags/https://humanityfirstresearch.com/tags/ai-safety/https://humanityfirstresearch.com/tags/apart-hackathon/https://humanityfirstresearch.com/tags/auditing/https://humanityfirstresearch.com/tags/backdoors/https://humanityfirstresearch.com/tags/behavioral/https://humanityfirstresearch.com/tags/boolq/https://humanityfirstresearch.com/tags/calibration/https://humanityfirstresearch.com/tags/causal-intervention/https://humanityfirstresearch.com/tags/chain-of-thought/https://humanityfirstresearch.com/tags/claudius-maximus-v0-01/https://humanityfirstresearch.com/tags/claudius-maximus-v0-02/https://humanityfirstresearch.com/tags/claudius-maximus-v0-03/https://humanityfirstresearch.com/tags/claudius-maximus-v0-04/https://humanityfirstresearch.com/tags/claudius-maximus-v0-05/https://humanityfirstresearch.com/tags/claudius-maximus-v0-06/https://humanityfirstresearch.com/tags/claudius-maximus-v0-07/https://humanityfirstresearch.com/tags/claudius-maximus-v0-08/https://humanityfirstresearch.com/tags/claudius-maximus-v0-10/https://humanityfirstresearch.com/tags/claudius-maximus-v0-12/https://humanityfirstresearch.com/tags/claudius-maximus-v0-13/https://humanityfirstresearch.com/tags/claudius-maximus-v0-15/https://humanityfirstresearch.com/tags/claudius-maximus-v0-16/https://humanityfirstresearch.com/tags/claudius-maximus-v0-17/https://humanityfirstresearch.com/tags/claudius-maximus-v0-18/https://humanityfirstresearch.com/tags/claudius-maximus-v0-20/https://humanityfirstresearch.com/tags/claudius-maximus-v0-21/https://humanityfirstresearch.com/tags/claudius-maximus-v0-24/https://humanityfirstresearch.com/tags/claudius-maximus-v0-25/https://humanityfirstresearch.com/tags/claudius-maximus-v0-29/https://humanityfirstresearch.com/tags/claudius-maximus-v0-30/https://humanityfirstresearch.com/tags/claudius-maximus-v0-31/https://humanityfirstresearch.com/tags/claudius-maximus-v0-32/https://humanityfirstresearch.com/tags/claudius-maximus-v0-33/https://humanityfirstresearch.com/tags/claudius-maximus-v0-34/https://humanityfirstresearch.com/tags/claudius-maximus-v0-36/https://humanityfirstresearch.com/tags/claudius-maximus-v0-39/https://humanityfirstresearch.com/tags/claudius-maximus-v0/https://humanityfirstresearch.com/tags/confidence/https://humanityfirstresearch.com/tags/contrastive-probe/https://humanityfirstresearch.com/tags/cot-faithfulness/https://humanityfirstresearch.com/tags/cot-monitorability/https://humanityfirstresearch.com/tags/cross-format/https://humanityfirstresearch.com/tags/deception-detection/https://humanityfirstresearch.com/tags/deception-honesty/https://humanityfirstresearch.com/tags/deception/https://humanityfirstresearch.com/tags/decodability-vs-use/https://humanityfirstresearch.com/tags/elicitation/https://humanityfirstresearch.com/tags/error-awareness/https://humanityfirstresearch.com/tags/error-detection/https://humanityfirstresearch.com/tags/faithfulness/https://humanityfirstresearch.com/tags/finetuning/https://humanityfirstresearch.com/tags/grpo/https://humanityfirstresearch.com/tags/hidden-states/https://humanityfirstresearch.com/tags/honesty/https://humanityfirstresearch.com/tags/interpretability/https://humanityfirstresearch.com/tags/knowledge-distillation/https://humanityfirstresearch.com/tags/logit-lens/https://humanityfirstresearch.com/tags/logit-transform/https://humanityfirstresearch.com/tags/mass-mean-probe/https://humanityfirstresearch.com/tags/mmlu/https://humanityfirstresearch.com/tags/model-generalization/https://humanityfirstresearch.com/tags/model-organisms/https://humanityfirstresearch.com/tags/monitorability/https://humanityfirstresearch.com/tags/null-result/https://humanityfirstresearch.com/tags/out-of-domain-generalization/https://humanityfirstresearch.com/tags/outcome-reward/https://humanityfirstresearch.com/tags/pass-at-k/https://humanityfirstresearch.com/tags/probing/https://humanityfirstresearch.com/tags/procrustes/https://humanityfirstresearch.com/tags/reinforcement-learning/https://humanityfirstresearch.com/tags/replication/https://humanityfirstresearch.com/tags/representation-geometry/https://humanityfirstresearch.com/tags/retraction/https://humanityfirstresearch.com/tags/rl-fine-tuning/https://humanityfirstresearch.com/tags/rl/https://humanityfirstresearch.com/tags/sandbagging/https://humanityfirstresearch.com/tags/scalable-oversight/https://humanityfirstresearch.com/tags/secret-loyalties/https://humanityfirstresearch.com/tags/self-consistency/https://humanityfirstresearch.com/tags/soft-labels/https://humanityfirstresearch.com/tags/sparse-autoencoders/https://humanityfirstresearch.com/tags/steganography/https://humanityfirstresearch.com/tags/sycophancy/https://humanityfirstresearch.com/tags/timing/https://humanityfirstresearch.com/tags/transfer/https://humanityfirstresearch.com/tags/tuned-lens/https://humanityfirstresearch.com/tags/weak-to-strong/https://humanityfirstresearch.com/tags/workshop-paper/https://humanityfirstresearch.com/team/