Symbolon Labs

Latent space / Interpretability

Latest result // R.01

A trained bridge between frozen language models.

A small trained bridge lets a frozen language model pass messages to a frozen model of another family. The first design failed a test of matching IDs to colors; a factored code later passed it on 56 unseen IDs (93.8%).

Read the R.01 entry

Research record

  1. R.01Latent channelIn progress
  2. R.02Receiver-native controllabilityNo-go
  3. R.03Residual-stream injectionNo-go
  4. R.04Cross-family linear mapEarly
  5. R.05Iterative latent exchangeIdea
  6. R.06Binding failure mechanismNo-go
  7. R.07Latent channel robustnessEarly
  8. R.08Uncertainty carrierIn progress
  9. R.09Split learning at equal bitsIn progress
  10. R.10Replay in learned message channelsEarly

Method

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

The full method
  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]