{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NJMH7MRSU4UYKWVTWJ5PROJIUP","short_pith_number":"pith:NJMH7MRS","schema_version":"1.0","canonical_sha256":"6a587fb232a729855ab3b27af8b928a3c6727a1e053c7275624bf9f53d2661e8","source":{"kind":"arxiv","id":"2512.10931","version":3},"attestation_state":"computed","paper":{"title":"Asynchronous Reasoning: Training-Free Interactive Thinking LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Modifying positional embeddings lets existing LLMs reason asynchronously while generating responses without any retraining.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Alina Shutova, Denis Kuznedelev, George Yakushev, Masoud Vahid Dastgerdi, Max Ryabinin, Nataliia Babina, Vyacheslav Zhdanovskiy","submitted_at":"2025-12-11T18:57:02Z","abstract_excerpt":"Many state-of-the-art LLMs are trained to think before giving their answer. Reasoning can greatly improve language model capabilities, but it also makes them less interactive: given a new input, a model must stop thinking before it can respond. Real-world use cases such as voice-based or embodied assistants require an LLM agent to respond and adapt to additional information in real time, which is incompatible with sequential interactions. In contrast, humans can listen, think, and act asynchronously: we begin thinking about the problem while reading it and continue thinking while formulating t"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2512.10931","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-12-11T18:57:02Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"4f49c30846d3d9d15e56db3b5e182fc1fbb8138c43807cb778c6f18344a684e6","abstract_canon_sha256":"b0de9e9afd4ce6f23c4c3c74d3b516a2839e93da22a6d344043a11ac1e688fd4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:32.173119Z","signature_b64":"lN2CypHVJvtcnfztq8aewzBS7U6p+PAVNRFvODn/5GXsEqxoFmNqlr01CLki5I7s3VZraipXryvxYV+x1lA3Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a587fb232a729855ab3b27af8b928a3c6727a1e053c7275624bf9f53d2661e8","last_reissued_at":"2026-05-18T02:44:32.172628Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:32.172628Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Asynchronous Reasoning: Training-Free Interactive Thinking LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Modifying positional embeddings lets existing LLMs reason asynchronously while generating responses without any retraining.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Alina Shutova, Denis Kuznedelev, George Yakushev, Masoud Vahid Dastgerdi, Max Ryabinin, Nataliia Babina, Vyacheslav Zhdanovskiy","submitted_at":"2025-12-11T18:57:02Z","abstract_excerpt":"Many state-of-the-art LLMs are trained to think before giving their answer. Reasoning can greatly improve language model capabilities, but it also makes them less interactive: given a new input, a model must stop thinking before it can respond. Real-world use cases such as voice-based or embodied assistants require an LLM agent to respond and adapt to additional information in real time, which is incompatible with sequential interactions. In contrast, humans can listen, think, and act asynchronously: we begin thinking about the problem while reading it and continue thinking while formulating t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our method uses the properties of positional embeddings to enable LLMs built for sequential generation to simultaneously think, listen, and write outputs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That altering the handling of positional embeddings preserves the model's original reasoning accuracy and does not introduce generation inconsistencies or hallucinations.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Using properties of positional embeddings, reasoning LLMs can be made to think, listen, and generate outputs asynchronously without any additional training, cutting time to first token to under 5 seconds.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Modifying positional embeddings lets existing LLMs reason asynchronously while generating responses without any retraining.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"50e92c2aa47d913bddd3b20ff285df87e88f26bdb9f0d23ccbd7109b8d6b359d"},"source":{"id":"2512.10931","kind":"arxiv","version":3},"verdict":{"id":"908e6e09-aa9c-4ea8-9afe-6126f1151aab","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T22:52:19.927882Z","strongest_claim":"Our method uses the properties of positional embeddings to enable LLMs built for sequential generation to simultaneously think, listen, and write outputs.","one_line_summary":"Using properties of positional embeddings, reasoning LLMs can be made to think, listen, and generate outputs asynchronously without any additional training, cutting time to first token to under 5 seconds.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That altering the handling of positional embeddings preserves the model's original reasoning accuracy and does not introduce generation inconsistencies or hallucinations.","pith_extraction_headline":"Modifying positional embeddings lets existing LLMs reason asynchronously while generating responses without any retraining."},"references":{"count":28,"sample":[{"doi":"10.1016/j.mlwa.2024.100570","year":2024,"title":"Beeching, E., Tunstall, L., and Rush, S","work_id":"b7cd3773-6405-41cb-a968-d32f1d014d2a","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1121/1.1906946","year":2023,"title":"Moshi: a speech-text foundation model for real-time dialogue","work_id":"3104332b-d279-44c8-aaa7-3d5a13c01832","ref_index":2,"cited_arxiv_id":"2410.00037","is_internal_anchor":true},{"doi":"10.1038/s41598-025-98378-1","year":2023,"title":"URL https: //doi.org/10.1038/s41598-025-98378-1","work_id":"15ee0369-b605-45b1-8ecb-1172423f7a90","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1038/s41586-025-09422-z","year":2025,"title":"Nature645(8081), 633–638 (2025) https://doi.org/10.1038/s41586-025-09422-z","work_id":"9835b482-5032-4135-93dd-82a066677569","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.48550/arxiv.2501.14249","year":2025,"title":"Humanity's Last Exam","work_id":"59ea00d4-16a8-45e1-aafc-290a6f91d9f4","ref_index":5,"cited_arxiv_id":"2501.14249","is_internal_anchor":true}],"resolved_work":28,"snapshot_sha256":"ebb61bfe032159341131288a525701f97ab90cbb7e11970552aa58a0ed5d5232","internal_anchors":12},"formal_canon":{"evidence_count":2,"snapshot_sha256":"b76857f71ad6afbf19c8921678c706d9b4d5d045bd6c97b096cddb55d0a1d5d1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2512.10931","created_at":"2026-05-18T02:44:32.172705+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.10931v3","created_at":"2026-05-18T02:44:32.172705+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.10931","created_at":"2026-05-18T02:44:32.172705+00:00"},{"alias_kind":"pith_short_12","alias_value":"NJMH7MRSU4UY","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"NJMH7MRSU4UYKWVT","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"NJMH7MRS","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.13360","citing_title":"Speculative Interaction Agents: Building Real-Time Agents with Asynchronous I/O and Speculative Tool Calling","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13360","citing_title":"Speculative Interaction Agents: Building Real-Time Agents with Asynchronous I/O and Speculative Tool Calling","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP","json":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP.json","graph_json":"https://pith.science/api/pith-number/NJMH7MRSU4UYKWVTWJ5PROJIUP/graph.json","events_json":"https://pith.science/api/pith-number/NJMH7MRSU4UYKWVTWJ5PROJIUP/events.json","paper":"https://pith.science/paper/NJMH7MRS"},"agent_actions":{"view_html":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP","download_json":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP.json","view_paper":"https://pith.science/paper/NJMH7MRS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.10931&json=true","fetch_graph":"https://pith.science/api/pith-number/NJMH7MRSU4UYKWVTWJ5PROJIUP/graph.json","fetch_events":"https://pith.science/api/pith-number/NJMH7MRSU4UYKWVTWJ5PROJIUP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP/action/storage_attestation","attest_author":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP/action/author_attestation","sign_citation":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP/action/citation_signature","submit_replication":"https://pith.science/pith/NJMH7MRSU4UYKWVTWJ5PROJIUP/action/replication_record"}},"created_at":"2026-05-18T02:44:32.172705+00:00","updated_at":"2026-05-18T02:44:32.172705+00:00"}