Independent research // no affiliation

Symbolon Research

Symbolon
Research
Latent space
Interpretability

Independent research on latent-space communication in language models. Each result is published with its controls, preregistration and negative results.

Symbolon Research // Record rev.01

Experiments use frozen models, on a laptop or on rented cloud GPUs. Thresholds are preregistered before data collection, destructive controls run beside every positive condition, and failed gates stay on the record with the reason they failed.

The research record ↓Method ↓

Research record.

06 entries // status as of this revision
  1. R.01Latent channel[ IN PROGRESS ]GATES: 0-4 PASSED
    BINDING: V1 FAILED, E3 PASSED
    REV: 01
  2. R.02Receiver-native controllability[ NO-GO ]AUDIT: SECOND SCRIPT
    RERUNS: 252 EXACT
    QUALIFYING DOSES: 0 OF 5
  3. R.03Residual-stream injection[ NO-GO ]PHASES: C-E
    C, D: GATES FAILED
    E: MECHANICS INVALID
  4. R.04Cross-family linear map[ EARLY WIP ]PAIRS: 5, FAMILIES: 3
    CARDS: 90.0-98.5%
    PREREGISTRATION: NONE
  5. R.05Iterative latent exchange[ IDEA ]DESIGN: NONE YET
    GATES: NONE YET
    RUNS: NONE
  6. R.06Binding failure mechanism[ NO-GO ]ORACLE CODES: 32/32
    LEARNED CODES: NOT COMPOSED
    NOVEL IDS: 0/32
Method

Negative results are published.

A positive result is informative only if the controls could have refuted it. Controls are specified first.

  1. 01Preregister the gateThresholds, magnitudes and pass conditions are fixed before any data is collected.
  2. 02Run the destructive controlsDestructive controls, for example zero, random, shuffled or text-handoff arms, run beside each correct condition. Each entry lists the arms that ran.
  3. 03Audit independentlyHashes, doses and confidence intervals are re-derived by a second implementation.
  4. 04Publish negative resultsFailed gates and retracted numbers stay on the record, with the reason they failed.

Contact

[email protected]
FIG.01 // Halftone sweep