{"id":"ai-observatory","title":"The AI Observatory: A Public Measure of Real-World AI Use","authors":[{"given":"Shayne","family":"Longpre"},{"given":"Anka","family":"Reuel"},{"given":"Dayeon","family":"Ki"},{"given":"Cathy Mengying","family":"Fang"},{"given":"Megan","family":"Richards"},{"given":"Zhiping","family":"Zhang"},{"given":"Chuanyang","family":"Jin"},{"given":"Jenn","family":"Mickel"},{"given":"Cedric Deslandes","family":"Whitney"},{"given":"Ahmet","family":"Üstün"},{"given":"Niloofar","family":"Mireshghallah"},{"given":"Victor","family":"Ojewale"},{"given":"Ariel","family":"Lee"},{"given":"Alex (Sandy)","family":"Pentland"},{"given":"Tianshi","family":"Li"},{"given":"Yuntian","family":"Deng"},{"given":"Mykel","family":"Kochenderfer"},{"given":"Sara","family":"Hooker"},{"given":"Sanmi","family":"Koyejo"}],"first_public":"2026-08-18","date_note":"Public launch coverage is dated August 18, 2026. This establishes public availability by that date, not the first manuscript submission. Counts and methods describe the current public PDF checked November 22, 2026.","date_source":"https://www.washingtonpost.com/business/2026/08/18/ai-observatory-tries-reveal-how-people-really-use-technology/","year":2026,"venue":"NeurIPS 2026 Evaluations and Datasets Track","paper_url":"https://www.dataprovenance.org/ai_observatory.pdf","category":"behavior","relationship":"uses","dataset_version":"23,158 conversations and 85,633 turns across seven sources. The annotated WildChat sample contains 14,648 conversations, 50,152 turns, and 4,439 users, drawn from unfiltered collection data spanning April 2023–July 2025.","summary":"Introduces a public measurement platform and a 145-feature taxonomy for comparing real-world AI use across sources, models, and time. WildChat supplies longitudinal conversations and returning-user data; the study connects these to cross-source analyses of tasks, interaction patterns, and sensitive uses.","finding":"Usage patterns differ substantially across sources. Applying an occupational filter removes 47.9% of the studied conversations, excluding personal, social, cultural, and safety-relevant uses.","evidence_url":"https://www.dataprovenance.org/ai_observatory.pdf","evidence_note":"Sections 2–3 and Table 1: sources, sampling, annotation, and analysis. References cite WildChat and WildBench. The public annotation release does not redistribute raw conversations.","bibtex_key":"longpre2026aiobservatory","record_url":"https://yuntiandeng.com/wildchat/papers/ai-observatory/","relationship_label":"Uses WildChat","citation_title":"The AI Observatory: A Public Measure of Real-World AI Use","citation_type":"paper-conference","bibtex_type":"inproceedings","citation_url":"https://www.dataprovenance.org/ai_observatory.pdf","citation_year":2026,"citation_note":"NeurIPS 2026 citation follows the authors’ public annotation dataset card. The linked PDF is the public manuscript.","date_label":"Public availability documented by","date_source_label":"Dated launch coverage","container_title":"Proceedings of the 40th Conference on Neural Information Processing Systems","author_note":"Shayne Longpre, Anka Reuel, and Dayeon Ki are co-first authors.","dataset_releases":[{"action":"Read","name":"AI Observatory annotations and citation","url":"https://huggingface.co/datasets/aiobservatory/annotations"}],"dataset_release_summary":"The released resource contains annotations rather than raw conversations."}
