{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:XJIJEQJ44MTYBQE2ZJXPTRSTDC","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"ba72141cdedaed3fe8fc260214b46024b12ab3ea8afdd5b7b98e1f5ecfc19c6f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T06:15:32Z","title_canon_sha256":"02b8f238bb5de79e2caa4bb12872c45e676aa76eef55f0d07d474f89fcc6523d"},"schema_version":"1.0","source":{"id":"2605.13050","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13050","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13050v2","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13050","created_at":"2026-05-18T03:08:59Z"},{"alias_kind":"pith_short_12","alias_value":"XJIJEQJ44MTY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"XJIJEQJ44MTYBQE2","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"XJIJEQJ4","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:8b9b1a729c57841d638a7ca717ce857479371541b7aaa5ce627cf72e6ac480eb","target":"graph","created_at":"2026-05-18T03:08:59Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"when paired with a search-based training procedure that maintains and prunes multiple candidate contexts, active information seeking delivers consistent and substantial gains"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the external search tools return sufficiently accurate and relevant passages and that the pruning step can reliably discard noisy or misleading contexts without discarding useful ones."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Active information seeking via search tools, when combined with multi-candidate context pruning during training, produces consistent gains on translation, health, and reasoning tasks over naive tool addition or no-tool baselines."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Pairing search tools with multi-candidate pruning during context training produces consistent LLM gains without weight updates."}],"snapshot_sha256":"14cbee5f2a2ff58a577876a8117c76e6806a8fc7679274c864dd4b90351725ec"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Most existing large language models (LLMs) are expensive to adapt after deployment, especially when a task requires newly produced information or niche domain knowledge. Recent work has shown that, by manipulating and optimizing their context, LLMs can be tailored to downstream tasks without updating their weights. However, most existing methods remain closed-loop, relying solely on the model's intrinsic knowledge. In this paper, we equip these context optimizers with Wikipedia search and browser tools for active information seeking. We show that naively adding these tools to a standard sequen","authors_text":"Adhiguna Kuncoro, Arthur Szlam, Jiajun Shen, Lucio Dery, Marc'Aurelio Ranzato, Qixuan Feng, Zeyu Huang","cross_cats":["cs.AI"],"headline":"Pairing search tools with multi-candidate pruning during context training produces consistent LLM gains without weight updates.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T06:15:32Z","title":"Context Training with Active Information Seeking"},"references":{"count":67,"internal_anchors":8,"resolved_work":67,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Mastering the game of","work_id":"9fd4a21f-922c-474a-8cbf-44afe2d81f74","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Trace is the Next AutoDiff: Generative Optimization with Rich Feedback, Execution Traces, and LLMs , author=. 2024 , eprint=","work_id":"bc1453c3-a276-46d7-b804-d5234277cd7f","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Arora and Jason Wei and Rebecca Soskin Hicks and Preston Bowman and Joaquin Qui","work_id":"5d0c0338-1729-4965-bffe-2e87858eb6ac","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Tom B. Brown and Benjamin Mann and Nick Ryder and Melanie Subbiah and Jared Kaplan and Prafulla Dhariwal and Arvind Neelakantan and Pranav Shyam and Girish Sastry and Amanda Askell and Sandhini Agarwa","work_id":"15cd97b7-6e24-48b5-b218-f433fead09cd","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Mondal and Jyoti Prakash Sahoo , title =","work_id":"9db2aa81-eda7-43c4-b26f-9b80ca6bbf44","year":null}],"snapshot_sha256":"8f67e2c38e8979cba2c7feccdd8e65ffd54ef867117ff3d82d7b2557e8d65c92"},"source":{"id":"2605.13050","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-15T06:03:28.957434Z","id":"74c46795-9804-419e-8625-b0f535493fb3","model_set":{"reader":"grok-4.3"},"one_line_summary":"Active information seeking via search tools, when combined with multi-candidate context pruning during training, produces consistent gains on translation, health, and reasoning tasks over naive tool addition or no-tool baselines.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Pairing search tools with multi-candidate pruning during context training produces consistent LLM gains without weight updates.","strongest_claim":"when paired with a search-based training procedure that maintains and prunes multiple candidate contexts, active information seeking delivers consistent and substantial gains","weakest_assumption":"That the external search tools return sufficiently accurate and relevant passages and that the pruning step can reliably discard noisy or misleading contexts without discarding useful ones."}},"verdict_id":"74c46795-9804-419e-8625-b0f535493fb3"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:ed86a2c791db7d03c32098dc587e9c15ca53ddab688d202428e6be044941c94e","target":"record","created_at":"2026-05-18T03:08:59Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"ba72141cdedaed3fe8fc260214b46024b12ab3ea8afdd5b7b98e1f5ecfc19c6f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T06:15:32Z","title_canon_sha256":"02b8f238bb5de79e2caa4bb12872c45e676aa76eef55f0d07d474f89fcc6523d"},"schema_version":"1.0","source":{"id":"2605.13050","kind":"arxiv","version":2}},"canonical_sha256":"ba5092413ce32780c09aca6ef9c6531894b166e0eba24a62102e2415e73c8ff5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ba5092413ce32780c09aca6ef9c6531894b166e0eba24a62102e2415e73c8ff5","first_computed_at":"2026-05-18T03:08:59.334343Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:08:59.334343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ocsN5K6URzevWudSt9jihI+z+hRODTz/I0d6r2p9nnJMr7HNlN1JGZe42NntCsLBlrgpC9FjD5ZSHwAG7z/kAw==","signature_status":"signed_v1","signed_at":"2026-05-18T03:08:59.336150Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13050","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:ed86a2c791db7d03c32098dc587e9c15ca53ddab688d202428e6be044941c94e","sha256:8b9b1a729c57841d638a7ca717ce857479371541b7aaa5ce627cf72e6ac480eb"],"state_sha256":"fee579cfbec290d4a0d61001ba7dbc0d57bdbefe3378b7ad00fe93adbc179114"}