{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:RIOFC5YCNUQFVWP4ZIXZVXDCLN","short_pith_number":"pith:RIOFC5YC","schema_version":"1.0","canonical_sha256":"8a1c5177026d205ad9fcca2f9adc625b5e01de8a7b5eb9639d13e134f0c9e03d","source":{"kind":"arxiv","id":"2310.14540","version":3},"attestation_state":"computed","paper":{"title":"Evaluating Spatial Understanding of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Andrew K. Lampinen, Ilker Yildirim, Jungo Kasai, Yihan Bao, Yutaro Yamada","submitted_at":"2023-10-23T03:44:40Z","abstract_excerpt":"Large language models (LLMs) show remarkable capabilities across a variety of tasks. Despite the models only seeing text in training, several recent studies suggest that LLM representations implicitly capture aspects of the underlying grounded concepts. Here, we explore LLM representations of a particularly salient kind of grounded knowledge -- spatial relationships. We design natural-language navigation tasks and evaluate the ability of LLMs, in particular GPT-3.5-turbo, GPT-4, and Llama2 series models, to represent and reason about spatial structures. These tasks reveal substantial variabili"},"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":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2310.14540","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-23T03:44:40Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"709a132563b05f55d3ba7ab5a1c1712633b020af6f2b0276529b3bf0636f9693","abstract_canon_sha256":"90c0f925beb7fc1eabe9155723c4c04b2e2259da810ae26198238d56ed2fac2d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:07:30.327476Z","signature_b64":"x/Fiz7Sozxy+N3ZQg/0yuNBYxnR2cHCn67GGggMTMv6phAFjjk/9wNS9/boGy674lUMQTvenU+lMkJjX8Sv/AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8a1c5177026d205ad9fcca2f9adc625b5e01de8a7b5eb9639d13e134f0c9e03d","last_reissued_at":"2026-07-05T08:07:30.326935Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:07:30.326935Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Evaluating Spatial Understanding of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Andrew K. Lampinen, Ilker Yildirim, Jungo Kasai, Yihan Bao, Yutaro Yamada","submitted_at":"2023-10-23T03:44:40Z","abstract_excerpt":"Large language models (LLMs) show remarkable capabilities across a variety of tasks. Despite the models only seeing text in training, several recent studies suggest that LLM representations implicitly capture aspects of the underlying grounded concepts. Here, we explore LLM representations of a particularly salient kind of grounded knowledge -- spatial relationships. We design natural-language navigation tasks and evaluate the ability of LLMs, in particular GPT-3.5-turbo, GPT-4, and Llama2 series models, to represent and reason about spatial structures. These tasks reveal substantial variabili"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.14540","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2310.14540/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2310.14540","created_at":"2026-07-05T08:07:30.327005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.14540v3","created_at":"2026-07-05T08:07:30.327005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.14540","created_at":"2026-07-05T08:07:30.327005+00:00"},{"alias_kind":"pith_short_12","alias_value":"RIOFC5YCNUQF","created_at":"2026-07-05T08:07:30.327005+00:00"},{"alias_kind":"pith_short_16","alias_value":"RIOFC5YCNUQFVWP4","created_at":"2026-07-05T08:07:30.327005+00:00"},{"alias_kind":"pith_short_8","alias_value":"RIOFC5YC","created_at":"2026-07-05T08:07:30.327005+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27378","citing_title":"Formalizing Latent Thoughts: Four Axioms of Thought Representation in LLMs","ref_index":82,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02211","citing_title":"Consistency Training while Mitigating Obfuscation via Rate Matching","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2503.19786","citing_title":"Gemma 3 Technical Report","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18380","citing_title":"QSTRBench: a New Benchmark to Evaluate the Ability of Language Models to Reason with Qualitative Spatial and Temporal Calculi","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2506.09965","citing_title":"Reinforcing Spatial Reasoning in Vision-Language Models with Interwoven Thinking and Visual Drawing","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2602.04850","citing_title":"El Agente Quntur: A research collaborator agent for quantum chemistry","ref_index":105,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN","json":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN.json","graph_json":"https://pith.science/api/pith-number/RIOFC5YCNUQFVWP4ZIXZVXDCLN/graph.json","events_json":"https://pith.science/api/pith-number/RIOFC5YCNUQFVWP4ZIXZVXDCLN/events.json","paper":"https://pith.science/paper/RIOFC5YC"},"agent_actions":{"view_html":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN","download_json":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN.json","view_paper":"https://pith.science/paper/RIOFC5YC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.14540&json=true","fetch_graph":"https://pith.science/api/pith-number/RIOFC5YCNUQFVWP4ZIXZVXDCLN/graph.json","fetch_events":"https://pith.science/api/pith-number/RIOFC5YCNUQFVWP4ZIXZVXDCLN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN/action/storage_attestation","attest_author":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN/action/author_attestation","sign_citation":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN/action/citation_signature","submit_replication":"https://pith.science/pith/RIOFC5YCNUQFVWP4ZIXZVXDCLN/action/replication_record"}},"created_at":"2026-07-05T08:07:30.327005+00:00","updated_at":"2026-07-05T08:07:30.327005+00:00"}