{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:AUOJIZBDOEHP76PESV763RVUXP","short_pith_number":"pith:AUOJIZBD","schema_version":"1.0","canonical_sha256":"051c946423710efff9e4957fedc6b4bbf5bebd1643b693362b19f1ddf469192b","source":{"kind":"arxiv","id":"2305.03495","version":2},"attestation_state":"computed","paper":{"title":"Automatic Prompt Optimization with \"Gradient Descent\" and Beam Search","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chenguang Zhu, Dan Iter, Jerry Li, Michael Zeng, Reid Pryzant, Yin Tat Lee","submitted_at":"2023-05-04T15:15:22Z","abstract_excerpt":"Large Language Models (LLMs) have shown impressive performance as general purpose agents, but their abilities remain highly dependent on prompts which are hand written with onerous trial-and-error effort. We propose a simple and nonparametric solution to this problem, Automatic Prompt Optimization (APO), which is inspired by numerical gradient descent to automatically improve prompts, assuming access to training data and an LLM API. The algorithm uses minibatches of data to form natural language \"gradients\" that criticize the current prompt. The gradients are then \"propagated\" into the prompt "},"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":"2305.03495","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-04T15:15:22Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"959bf99b0b128ac81d9186058b4c9df323295a0a2338b928e96c0be4e7ee4e16","abstract_canon_sha256":"03bb66f3e9d1dbd00fbe5ad028e839723a51484d33c425e15afffedcd8a85d17"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:02:28.636917Z","signature_b64":"d0mDExlai2/6b0XeKQtmVArXS0n7TA2G/S1inhVkASCFh5/KqQSzwn8a5fuix+/SuVQNFBTn733L5M8zouf/Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"051c946423710efff9e4957fedc6b4bbf5bebd1643b693362b19f1ddf469192b","last_reissued_at":"2026-07-05T07:02:28.636416Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:02:28.636416Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Automatic Prompt Optimization with \"Gradient Descent\" and Beam Search","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chenguang Zhu, Dan Iter, Jerry Li, Michael Zeng, Reid Pryzant, Yin Tat Lee","submitted_at":"2023-05-04T15:15:22Z","abstract_excerpt":"Large Language Models (LLMs) have shown impressive performance as general purpose agents, but their abilities remain highly dependent on prompts which are hand written with onerous trial-and-error effort. We propose a simple and nonparametric solution to this problem, Automatic Prompt Optimization (APO), which is inspired by numerical gradient descent to automatically improve prompts, assuming access to training data and an LLM API. The algorithm uses minibatches of data to form natural language \"gradients\" that criticize the current prompt. The gradients are then \"propagated\" into the prompt "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.03495","kind":"arxiv","version":2},"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/2305.03495/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":"2305.03495","created_at":"2026-07-05T07:02:28.636478+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.03495v2","created_at":"2026-07-05T07:02:28.636478+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.03495","created_at":"2026-07-05T07:02:28.636478+00:00"},{"alias_kind":"pith_short_12","alias_value":"AUOJIZBDOEHP","created_at":"2026-07-05T07:02:28.636478+00:00"},{"alias_kind":"pith_short_16","alias_value":"AUOJIZBDOEHP76PE","created_at":"2026-07-05T07:02:28.636478+00:00"},{"alias_kind":"pith_short_8","alias_value":"AUOJIZBD","created_at":"2026-07-05T07:02:28.636478+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":23,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25193","citing_title":"LLM4MTLs: Automated Generation and Empirical Evaluation of Model Transformation Languages","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22803","citing_title":"Towards Fast Domain Adaptation and Fine-Grained User Simulation for Evaluating Conversational Recommender Systems","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22385","citing_title":"MetaPS: Adaptive Programmatic Strategy Selection for Market Agents","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20475","citing_title":"Marginal Advantage Accumulation for Memory-Driven Agent Self-Evolution","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18620","citing_title":"BCL: Bayesian In-Context Learning Framework for Information Extraction","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12387","citing_title":"TAHOE: Text-to-SQL with Automated Hint Optimization from Experience","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08867","citing_title":"Building Customer Support AI Agents at 100M-User Scale: An Evaluation-Driven Framework","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00427","citing_title":"BT-APE: A Computationally Light Backtracking Approach to Automatic Prompt Engineering for Requirements Classification","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2603.06610","citing_title":"CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18780","citing_title":"A Reproducibility Analysis of PO4ISR: Diagnosing and Mitigating Semantic Drift in LLM-Based Session Recommendation","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15803","citing_title":"Embedding-perturbed Exploration Preference Optimization for Flow Models","ref_index":68,"is_internal_anchor":false},{"citing_arxiv_id":"2310.11324","citing_title":"Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2512.11013","citing_title":"PIAST: Rapid Prompting with In-context Augmentation for Scarce Training data","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2309.16797","citing_title":"Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2309.08532","citing_title":"EvoPrompt: Connecting LLMs with Evolutionary Algorithms Yields Powerful Prompt Optimizers","ref_index":124,"is_internal_anchor":false},{"citing_arxiv_id":"2603.01692","citing_title":"Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2309.03409","citing_title":"Large Language Models as Optimizers","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2507.21046","citing_title":"A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2603.28052","citing_title":"Meta-Harness: End-to-End Optimization of Model Harnesses","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2310.03714","citing_title":"DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08224","citing_title":"Externalization in LLM Agents: A Unified Review of Memory, Skills, Protocols and Harness Engineering","ref_index":120,"is_internal_anchor":false},{"citing_arxiv_id":"2303.11366","citing_title":"Reflexion: Language Agents with Verbal Reinforcement Learning","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14585","citing_title":"Prompt Optimization Is a Coin Flip: Diagnosing When It Helps in Compound AI Systems","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP","json":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP.json","graph_json":"https://pith.science/api/pith-number/AUOJIZBDOEHP76PESV763RVUXP/graph.json","events_json":"https://pith.science/api/pith-number/AUOJIZBDOEHP76PESV763RVUXP/events.json","paper":"https://pith.science/paper/AUOJIZBD"},"agent_actions":{"view_html":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP","download_json":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP.json","view_paper":"https://pith.science/paper/AUOJIZBD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.03495&json=true","fetch_graph":"https://pith.science/api/pith-number/AUOJIZBDOEHP76PESV763RVUXP/graph.json","fetch_events":"https://pith.science/api/pith-number/AUOJIZBDOEHP76PESV763RVUXP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP/action/storage_attestation","attest_author":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP/action/author_attestation","sign_citation":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP/action/citation_signature","submit_replication":"https://pith.science/pith/AUOJIZBDOEHP76PESV763RVUXP/action/replication_record"}},"created_at":"2026-07-05T07:02:28.636478+00:00","updated_at":"2026-07-05T07:02:28.636478+00:00"}