{"event":{"id":"evt-toolformer-20230209","dedupe_key":"tool_use_research:2023-02-09:toolformer-demonstrates-language-model-api-tool-use","event_type":"tool_use_research","title":"Toolformer demonstrates language-model API tool use","summary":"Toolformer showed a language model learning when and how to call external tools such as calculators, search and translation APIs.","occurred_at":"2023-02-09T00:00:00.000Z","published_at":"2023-02-09T00:00:00.000Z","observed_at":"2026-10-05T14:55:19.000Z","reconstructed_at":"2026-10-05T15:23:51.000Z","ingest_type":"live_world_scan","locations":null,"status":"verified","confidence":0.98,"metadata":{"actors":["arXiv"],"what_happened_at_time":"Toolformer showed a language model learning when and how to call external tools such as calculators, search and translation APIs.","evidence_at_time":"The February 9 Toolformer preprint by Timo Schick et al. reports experiments in which a language model learns when and how to invoke tools including a calculator, search, translation, and calendar-style APIs.","affected_layers":["layer-models","layer-agents","layer-research","layer-software"],"change_kind":"tool_use_research_maturation","q4_continuity":"Q4 is the canonical predecessor; this record stores only the supported Q1 delta.","retrospective_inference":"Interpretation is limited to Q1 evidence and does not import later 2023-2026 consequences.","uncertainty":"Access, deployment, adoption and downstream effects are kept distinct from the announcement or observed capability."},"created_at":"2026-10-05T18:17:53.787Z","updated_at":"2026-10-05T18:17:53.787Z"},"artifacts":[{"id":"art-toolformer-20230209","canonical_url":"https://arxiv.org/abs/2302.04761","artifact_type":"tool_use_research","title":"Toolformer demonstrates language-model API tool use","summary":"Toolformer showed a language model learning when and how to call external tools such as calculators, search and translation APIs.","creator_entities":["Timo Schick et al."],"released_at":"2023-02-09T00:00:00.000Z","content_hash":null,"metadata":{"historical_scope":"2023-Q1","creator_note":"arXiv is the publication platform; creator attribution follows the source author field."},"created_at":"2026-10-05T18:17:53.780Z","updated_at":"2026-10-05T18:17:53.780Z","relation":"evidenced_by"}],"signals":[{"id":"sig-q1-agent-architecture-expansion","statement":"Research and bounded products combined reasoning, API/tool use, external information and goal-directed action more explicitly, while still falling short of persistent general-purpose autonomous agents.","signal_type":"agent_architecture_maturation","direction":"increasing","confidence":0.92,"epistemic_status":"supported_inference","mapping_mode":"retrospective","reconstructed_at":"2026-10-05T15:23:51.000Z","created_at":"2026-10-05T18:17:53.796Z","updated_at":"2026-10-05T18:17:53.796Z"},{"id":"sig-q1-tool-use-productization","statement":"Tool use moved from research-only evidence toward constrained product deployment through ChatGPT plugins, including browsing, code execution and third-party service interaction.","signal_type":"tool_use_productization","direction":"increasing","confidence":0.96,"epistemic_status":"supported_inference","mapping_mode":"retrospective","reconstructed_at":"2026-10-05T15:23:51.000Z","created_at":"2026-10-05T18:17:53.795Z","updated_at":"2026-10-05T18:17:53.795Z"}]}