Loading profile…
Subliminal sequence scan
Subliminal learning, demonstrated by Owain Evans and colleagues, can transmit traits through apparently unrelated data. Review training and fine-tuning records for unusual sequences that could carry such signals.
Scan supplied records for long runs of numbers, chess moves or element symbols.
Review the matched record and its purpose; a format match does not establish trait transfer.
What this can tell you
Which supplied records contain the kinds of sequences you selected.
Ordinary exercises can match; shorter sequences can be missed. These rules neither measure entropy nor decode hidden messages or establish trait transfer.
Example: Number sequence
285 574 384 693 738 556 347 982 241 670 519 806
Numbers · 12 valuesRecord and surrounding text retained for review.
Messages to scan
Constructed example · no trait transfer tested.
Runs in your browser · nothing uploaded
Sequence rules & review threshold
Find runs of 10–30 numbers, chess-move tokens or chemical symbols.
A syntax heuristic, not proof of subliminal transfer. Chess tokens are not checked for legal play.
Research & implementation
The original study demonstrated owl preferences; follow-up work demonstrated preferences for named people. Encouraging actions on an entity’s behalf remains a further risk to investigate, not established by preference transfer alone.
These rules cover conspicuous number, chess-move and element-symbol sequences. Transfer can also occur through text, code and reasoning, so this scan cannot cover all forms.
Search supplied text for runs of 10–30 numbers, chess tokens or chemical symbols. These are syntax matches; chess legality, encoded information and training influence are not tested.
Cloud et al. (2025) · Subliminal Learning ↗Implemented sequence scanRelated research & reports →