bamdadd/introspection-scaling

Cross-architecture probe: GLM family at the pinned dose 14 days ago

help wanted type: research area: llm size: L

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Cross-architecture probe: Kimi (MoE) at the pinned dose 14 days ago

help wanted type: research area: llm size: L

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Cross-architecture probe: does the code-post-training introspection effect appear in a MoE? 16 days ago

help wanted

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Expose --max-new-tokens on the runner CLI 19 days ago

help wanted good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Cross-architecture probe: introspection on a small open MoE 19 days ago

help wanted

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

7B/14B base-vs-instruct-vs-coder triple (turn the dissociation into a trend) 19 days ago

help wanted

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Add unit tests for resolve_dose (both dose modes + error paths) 20 days ago

good first issue type: test area: infra size: S

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Prototype a VLM concept-injection introspection probe 20 days ago

help wanted type: research area: vlm size: L

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Logit-lens layer sweep: where does the injected concept become legible? 20 days ago

help wanted type: research area: interp size: M

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Wire the deferred Llama-3.x rungs at the corrected dose 20 days ago

help wanted type: feature area: llm size: M

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Add a base-model rung for base-vs-instruct comparison 20 days ago

help wanted type: feature area: llm size: M

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Default dose-mode is the superseded under-dosed resid_frac 20 days ago

help wanted type: bug area: interp size: S

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Update stale α=0.044·‖resid‖ dose references in the README body 20 days ago

good first issue type: docs area: interp size: S

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Add unit tests for _looks_coherent (criterion-1 coherence heuristic) 22 days ago

help wanted good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Document the API-key-free (RuleBasedJudge) smoke path in the Quickstart 22 days ago

help wanted good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Decouple GPU generation from API judging (cut GPU cost + wall-clock) 23 days ago

enhancement good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Make BASELINE_WORDS pluggable via a --baseline-file CLI flag 23 days ago

enhancement good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Implement the paper's trial-scoped injection span in RepengGenerator 23 days ago

enhancement good first issue

bamdadd/introspection-scaling

0

Does concept-injection introspection emerge with scale? A faithful, controlled reproduction charted across model-size ladders.

Python

Open Issues Need Help