{"paper":{"title":"Borrowed Geometry: Cross-Distribution Head-Importance Fingerprints of Frozen Pretrained Gemma 4 31B","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Frozen text-pretrained transformer weights transfer to robotic and memory tasks through a thin trainable interface without modifying the core model.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Abay Bektursun","submitted_at":"2026-05-01T01:23:37Z","abstract_excerpt":"Frozen Gemma 4 31B weights pretrained exclusively on text, unmodified, transfer through a thin trainable interface to non-text modalities the substrate has never processed. On the L24--L29 slice (192 attention heads), an English-text TxtCopy attention probe (95 sentences) and per-head ablation impact on four non-language token-pattern tasks (binary copy, associative recall, 1D cellular automaton Rule 90, binary addition) jointly classify four heads -- L26.28, L27.28, L27.2, L27.3 -- as top-tier on both signals. The slice-level joint coincidence is significant under hypergeometric null ($P = 0."},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Frozen Gemma 4 31B weights pretrained exclusively on text tokens, unmodified, transfer across modality boundaries through a thin trainable interface.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the geometry captured by text-only pretraining remains useful and transferable to non-text modalities without any modification to the core frozen weights.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Frozen text-pretrained transformer weights transfer across modalities through a thin interface, achieving SOTA on a robotic task and parity on decision-making with far fewer trainable parameters.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Frozen text-pretrained transformer weights transfer to robotic and memory tasks through a thin trainable interface without modifying the core model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"207760f4c8b4cafd86db2d9ab3f7eecedd68784e6906469ffbc03096cd85f416"},"source":{"id":"2605.00333","kind":"arxiv","version":2},"verdict":{"id":"05f7c2ff-cec2-49f0-a49c-b5617c649e24","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-09T20:27:37.081861Z","strongest_claim":"Frozen Gemma 4 31B weights pretrained exclusively on text tokens, unmodified, transfer across modality boundaries through a thin trainable interface.","one_line_summary":"Frozen text-pretrained transformer weights transfer across modalities through a thin interface, achieving SOTA on a robotic task and parity on decision-making with far fewer trainable parameters.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the geometry captured by text-only pretraining remains useful and transferable to non-text modalities without any modification to the core frozen weights.","pith_extraction_headline":"Frozen text-pretrained transformer weights transfer to robotic and memory tasks through a thin trainable interface without modifying the core model."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00333/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T18:15:56.124031Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"567a48c8ba3438466c5ca81a4a531495591bb5fbeddc58ec7ef5dfb71b0a80a6"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}