Can a learned Markov chain over a person's webcam frames fool a detector?

What it found

A first-order Markov chain over k-means states (k = 8, 32, 128) of DUX webcam-signal frames, fitted on one person's earlier windows or on the other 45 people's, emitting state centres or real frames, is scored against that person's later windows by a boosted-tree detector (3 seeds). Every cell is AUROC 1.000 except pooled 128-state member emission at 0.999, in the population and the profile-aware detector alike. Nearly every feature group alone separates them at 0.92 or more, so the chain gets none of the window-scale statistics right even when it emits real frames; 50-frame blocks of the same frames reached 0.57 to 0.65 in the predecessor project. A guess that chains lose between-window differences is only partly supported. Stylised: one form-filling context, first-order chains only.

A figure from its own output

Paper, code and results

• README and method
• Experiments and results
• Tests

More in Adversarial human imitation

• Does fitting a webcam-signal generator to one person fool a detector that has seen that person?Does fitting to one person help? Copying a person's own earlier windows is caught at AUROC 0.59 to 0.65 on their later windows; copying other people is 0.58 to 0.91, and 0.93 to 0.97 once the detector knows the person; a fitted parametric generator is caught at 1.000 either way.
• Does giving a Markov chain dwell times or a second-order memory let it fool a webcam-signal detector?Does giving a chain over a person's frames dwell times or second-order memory help? No: semi-Markov and second-order chains (8 or 32 states, own or pooled fit, centre or real-frame emission) are all caught at AUROC 0.999 to 1.000.
• Why is block-stitched real webcam signal still caught: the joins, or mixing people?Why is block-stitched real webcam signal still caught? With 50-frame blocks random stitching gives AUROC 0.887, one person's windows only 0.687, join-matched 0.796, both 0.588 (= whole-window replay 0.597); at 5 frames all are caught (8 seeds).
• How well do prompt-free detectors tell people from a plainly prompted LLM, and do they carry over to unseen kinds of text?How well do prompt-free detectors tell people from a plainly prompted LLM, and do they carry over to unseen kinds of text? Eight RAID families split by family: n-gram detector AUROC 0.93 and Binoculars-style 0.80 on held-out families, length at chance. Preliminary first step of track 11.
• Do the file sizes and line counts of this repository follow Benford’s law?Do this repository's file sizes follow Benford's law? Not as a whole (byte MAD 0.018, nonconforming) because templated files repeat the same size thousands of times; counting each distinct size once gives acceptable conformity (MAD 0.0068).
• Can prompt search alone move a frozen 0.5B imitator past a prompted 0.5B detector, and which proposer does it best at equal budget?Can prompt search alone move a frozen 0.5B imitator past a prompted 0.5B detector? A stylised first run of the imitation game, four proposers at equal budget on two families: the starting detector is below chance, proposers cannot be ranked at this size, and length is a major confound.

All 41 Adversarial human imitation projects →

Other research topics

Proper scoring & elicitation28Effort & peer prediction9Verification games & slashing19Detection & tolerances14Markets as routers13Prediction-market design9Paying contributors8Decentralised training systems23Digital twins & sim-to-real164LLM chip design loops40Robot manufacturing loops53Ambient AI assistant46Self-replicating manufacturing34Simulation-first design64Engineered clothing45Task-specific AI accelerators40Algorithmic chip-design loops42Open-source purpose-built robots44The software-engineering loop43Adversarial human imitation41Open-source residential construction41Autonomous multi-angle capture34Sensing, imaging and 3D31Energy supply and dense energy storage39Robot societies among people31Agentic co-design of low-cost robots27Laundry-hanging robots25Machine learning on qualitative data23Robots on the building site27Civilizations & group psychology1History & human sciences1

Produced with AI assistance (Claude). Results are from the code in this repository and are simulated or stylised unless the abstract says otherwise; nothing here implies affiliation with or endorsement by any lab, researcher or company cited. MIT licensed.