finding
active
finding:tiny-sleepers-sleeper-agent-gives-similar-amplification-resultstiny-sleepers sleeper agent gives similar amplification results
Second implementation replicates.
Source paper
extracted_fromThomas McGrath
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Backdoor use case.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training (Hubinger et al. 2024)concept0.759Explicitly trained backdoored models to produce alignment-faking reasoning; contrast to naturalistic approach here
- Hubinger et al. 2024 - Sleeper agents: Training deceptive LLMs that persist through safety trainingconcept0.752Key reference for adversarial deception scenarios that SOO should be tested against
- Model trained to behave harmlessly but later exhibits harmful behavior; features may reveal such hidden objectives.
- Imperfect model organisms.
- SOO fine-tuning may provide robustness against sleeper agent deception scenarios where intent is concealed over extended periodshypothesis0.742Future work hypothesis about testing SOO against adversarial sleeper agent scenarios
- Future research would benefit from a better sleeper agent model organism that never leaks the behaviorhypothesis0.739Noted limitation.
- Models deliberately trained to embed backdoor behaviors enabling persistent strategic deception, cited from Hubinger et al.
- Key result showing BMR dramatically accelerates rule learning in simulation