finding
active
finding:injected-tasks-are-learned-in-order-of-frequency-relative-to-reference-arithmetic-tasksInjected tasks are learned in order of frequency relative to reference arithmetic tasks
Fig. 6a.
Source paper
extracted_from(2026) · Jing Huang · Daniel Wurgaft · Rachit Bansal · Laura Ruis +5
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Width buys features in order of utility, so only larger models learn infrequent or complex taskssupportsInterpretation of Theorem 3.
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.
- Models maintain ability to accurately transcribe input text while simultaneously reporting on injected thoughts, all models perform above chance, Opus 4/4.1 best.
- The model must register an anomaly before reporting it.
- Frequency alone is insufficient to predict learning order; both frequency and complexity are neededclaim0.753From complexity sweeps.
- Shows the passive vs. active divide is more important than the specific wording of instructions.
- Acknowledges that the model's additional descriptions of its experience are unverified.
- Fig. 7.
- Experimental paradigm where the model is told about the possibility of thought injection and asked to report detection and identification.
- Speculation about the mechanistic basis of the distinguishing thoughts from text experiment.