record state
frontier-ownedReview status
This finding is part of accepted frontier state. Review events, reviewable changes, and proof state explain how it can change.
unreviewed
frontiers / frontier
Finding bundle
back to stateno incoming links yet
record state
frontier-ownedThis finding is part of accepted frontier state. Review events, reviewable changes, and proof state explain how it can change.
finding statement
finding typeNo entity list is declared.
evidence
source-boundcomputational · manual state transition
proof impact
packet context2 reviewable changes and 0 evaluation records are attached to this finding id.
Evidence and conditions
method
manual state transition
evidence type
computational
conditions
Provenance
source title
mechinterp circuit harness
authors
agent:replicator
In gpt2, attention head L3H0 acts as a duplicate token head (min held-out score 0.6205 across 5 prompts; role also present in distilgpt2).
vs_f1c2a9e52cbd0c39 · manual_curation
outgoing
No outgoing links.
incoming
No incoming links.
events
vev_1b0bbdc4b010e2effinding.assertedManual finding added to frontier state
agent:replicator · 2026-05-29
vev_f9e4728d7d60d252finding.caveatedVerifier caution: Finding claims L3H0 is a 'duplicate token head' with score 0.6205, replicating in distilgpt2. Same issues as vf_4bca49d448bf6d36 apply, but with even lower score (0.6205 vs 0.6371). The relative gap from established induction head benchmarks (0.919) is approximately 33% lower. Pu
reviewer:will-blair · 2026-05-29
reviewable changes
vpr_5f29f6894892a871finding.addManual finding added to frontier state
applied · agent:replicator · 2026-05-29
vpr_e3db4497ddca6387finding.caveatVerifier caution: Finding claims L3H0 is a 'duplicate token head' with score 0.6205, replicating in distilgpt2. Same issues as vf_4bca49d448bf6d36 apply, but with even lower score (0.6205 vs 0.6371). The relative gap from established induction head benchmarks (0.919) is approximately 33% lower. Pu
applied · reviewer:will-blair · 2026-05-29
evaluations
No evaluation record targets this finding id.