{"schema_version":"onlylabs.public_signal.v1","title":"OpenAI Writing: Teaching models to express their uncertainty in words","description":"OpenAI writing signal with public source context, captured evidence pages, related signals, and data-business radar classification.","url":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78","json_url":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78/signal.json","generated_at":"2026-06-08T15:46:48.454+00:00","org":{"slug":"openai","name":"OpenAI","category":"frontier-lab","category_label":"Frontier lab","dossier_url":"https://onlylabs.fyi/labs/openai","dossier_json_url":"https://onlylabs.fyi/labs/openai/dossier.json"},"related_urls":{"signal":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78","signal_json":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78/signal.json","source":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","lab_dossier":"https://onlylabs.fyi/labs/openai","lab_dossier_json":"https://onlylabs.fyi/labs/openai/dossier.json","analysis":"https://onlylabs.fyi/analysis/openai","analysis_json":"https://onlylabs.fyi/analysis/openai/analysis.json","analysis_evidence_json":"https://onlylabs.fyi/analysis/openai/evidence.json","category":"https://onlylabs.fyi/frontier","category_json":"https://onlylabs.fyi/frontier.json","category_feed":"https://onlylabs.fyi/frontier/feed.xml","category_signals_json":"https://onlylabs.fyi/signals.json","topic":"https://onlylabs.fyi/topics/talking","topic_signals_json":"https://onlylabs.fyi/topics/talking/signals.json","topic_feed":"https://onlylabs.fyi/topics/talking/feed.xml","data_business":null},"answer_pack":{"answer":"OpenAI published Teaching models to express their uncertainty in words. This talking signal gives public context for research themes, product direction, policy, or launch framing. High-signal details: Teaching models to express their uncertainty in words | OpenAI May 28, 2022 Teaching models to express their uncertainty in words Loading… Share Abstract We show that a.... onlylabs links this event to 1 captured evidence page and 6 related writing signals.","signal_desk":"talking","source_context":{"source_url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","source_host":"openai.com","occurred_at":"2022-05-28T07:00:00+00:00","first_seen_at":"2026-06-05T05:42:57.832854+00:00","date_source":"rss.item_date","context":null},"context_markers":[{"label":"Lab","value":"OpenAI","source":"signal"},{"label":"Signal desk","value":"talking","source":"signal"},{"label":"Source host","value":"openai.com","source":"source"}],"evidence_coverage":{"target_pages":1,"captured_pages":1,"readable_pages":1,"capture_methods":["exa"],"missing_page_urls":[],"failed_page_urls":[],"blocked_page_urls":[],"page_urls":["https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words"],"related_signals":6,"has_source_url":true,"latest_page_fetched_at":"2026-06-08T15:46:48.454+00:00"},"data_business":{"matches":false,"lanes":[],"matched_terms":[],"score":null,"reason":null},"agent_handoff":{"signal_json":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78/signal.json","dossier_json":"https://onlylabs.fyi/labs/openai/dossier.json","analysis_json":"https://onlylabs.fyi/analysis/openai/analysis.json","analysis_evidence_json":"https://onlylabs.fyi/analysis/openai/evidence.json","topic_signals_json":"https://onlylabs.fyi/topics/talking/signals.json","topic_feed":"https://onlylabs.fyi/topics/talking/feed.xml","category_signals_json":"https://onlylabs.fyi/signals.json","data_radar_json":null,"opportunities_json":null},"analysis_playbook":{"objective":"Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.","evidence_focus":["post title","source URL","captured page text","HN traction","linked model or paper references","publication date"],"extraction_questions":["Which themes are labs choosing to explain publicly?","Which posts are attracting outside discussion?","Which writing reframes a recent release, model, hiring wave, or policy stance?","Which posts mention data, evals, infrastructure, safety, or deployment workflows?"],"signal_questions":["What public theme, launch framing, or research direction does this writing signal expose?","Which themes are labs choosing to explain publicly?","Which posts are attracting outside discussion?","Do the 6 related writing signals show a repeated pattern?"],"output_fields":["org","theme","public_framing","traction","data_business_lane","evidence_url"],"data_business_relevance":"Public writing supplies the narrative layer over raw signals and helps identify which frontier-lab priorities are becoming externally legible.","required_sources":[{"label":"signal_json","url":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78/signal.json","required":true},{"label":"source","url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","required":true},{"label":"dossier_json","url":"https://onlylabs.fyi/labs/openai/dossier.json","required":true},{"label":"analysis_evidence_json","url":"https://onlylabs.fyi/analysis/openai/evidence.json","required":true},{"label":"topic_signals_json","url":"https://onlylabs.fyi/topics/talking/signals.json","required":false},{"label":"data_radar_json","url":null,"required":false}],"expected_output":["one-paragraph source-grounded interpretation","category-specific implication","confidence and missing evidence","recommended next source to inspect"],"prompt_seed":"Using only the linked onlylabs JSON, captured source context, and cited evidence, analyze OpenAI's writing signal \"Teaching models to express their uncertainty in words\" for frontier lab strategy."},"semantic_triples":[{"subject":"OpenAI","predicate":"published","object":"Teaching models to express their uncertainty in words","text":"OpenAI published Teaching models to express their uncertainty in words."},{"subject":"Teaching models to express their uncertainty in words","predicate":"is classified as","object":"writing signal","text":"Teaching models to express their uncertainty in words is classified as writing signal."},{"subject":"Teaching models to express their uncertainty in words","predicate":"belongs to","object":"talking desk","text":"Teaching models to express their uncertainty in words belongs to talking desk."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has evidence coverage","object":"1 captured evidence page","text":"Teaching models to express their uncertainty in words has evidence coverage 1 captured evidence page."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has captured page count","object":"1","text":"Teaching models to express their uncertainty in words has captured page count 1."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has readable page count","object":"1","text":"Teaching models to express their uncertainty in words has readable page count 1."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has related signal count","object":"6","text":"Teaching models to express their uncertainty in words has related signal count 6."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has analysis playbook objective","object":"Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.","text":"Teaching models to express their uncertainty in words has analysis playbook objective Turn public writing and discussion into a readable map of research themes, product framing, policy posture, launch narratives, and market attention.."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has source host","object":"openai.com","text":"Teaching models to express their uncertainty in words has source host openai.com."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has lab","object":"OpenAI","text":"Teaching models to express their uncertainty in words has lab OpenAI."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has signal desk","object":"talking","text":"Teaching models to express their uncertainty in words has signal desk talking."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has source host","object":"openai.com","text":"Teaching models to express their uncertainty in words has source host openai.com."}]},"intelligence":{"signal_desk":"talking","answer":"OpenAI published Teaching models to express their uncertainty in words. This talking signal gives public context for research themes, product direction, policy, or launch framing. High-signal details: Teaching models to express their uncertainty in words | OpenAI May 28, 2022 Teaching models to express their uncertainty in words Loading… Share Abstract We show that a.... onlylabs links this event to 1 captured evidence page and 6 related writing signals.","semantic_triples":[{"subject":"OpenAI","predicate":"published","object":"Teaching models to express their uncertainty in words","text":"OpenAI published Teaching models to express their uncertainty in words."},{"subject":"Teaching models to express their uncertainty in words","predicate":"is classified as","object":"writing signal","text":"Teaching models to express their uncertainty in words is classified as writing signal."},{"subject":"Teaching models to express their uncertainty in words","predicate":"belongs to","object":"talking desk","text":"Teaching models to express their uncertainty in words belongs to talking desk."},{"subject":"Teaching models to express their uncertainty in words","predicate":"has evidence coverage","object":"1 captured evidence page","text":"Teaching models to express their uncertainty in words has evidence coverage 1 captured evidence page."}]},"signal":{"id":"9321959c-3597-4d58-bee8-046cd3d37e78","url":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78","json_url":"https://onlylabs.fyi/signals/9321959c-3597-4d58-bee8-046cd3d37e78/signal.json","source_url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","title":"Teaching models to express their uncertainty in words","summary":"OpenAI published a writing signal. onlylabs watches public writing for research themes, product direction, and model-launch context.","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2022-05-28T07:00:00+00:00","first_seen_at":"2026-06-05T05:42:57.832854+00:00","date_source":"rss.item_date","evidence_coverage":{"target_pages":1,"captured_pages":1,"readable_pages":1,"capture_methods":["exa"],"missing_page_urls":[],"failed_page_urls":[],"blocked_page_urls":[],"page_urls":["https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words"]},"facets":{},"traction":{"github_stars":null,"hn_points":null,"hn_comments":null,"hn_story_id":null,"hf_downloads":null,"hf_likes":null},"data_radar":null},"primary_evidence_page":{"url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","final_url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","title":"Teaching models to express their uncertainty in words","http_status":200,"content_type":null,"capture_method":"exa","fetched_at":"2026-06-08T15:46:48.454+00:00","bytes":null,"raw_path":null,"content_hash":null,"excerpt_chars":1200,"truncated":true,"excerpt":"Teaching models to express their uncertainty in words | OpenAI May 28, 2022 Teaching models to express their uncertainty in words Loading… Share Abstract We show that a GPT‑3 model can learn to express uncertainty about its own answers in natural language—without use of model logits. When given a question, the model generates both an answer and a level of confidence (e.g. \"90% confidence\" or \"high confidence\"). These levels map to probabilities that are well calibrated. The model also remains moderately calibrated under distribution shift, and is sensitive to uncertainty in its own answers, rather than imitating human examples. To our knowledge, this is the first time a model has been shown to express calibrated uncertainty about its own answers in natural language. For testing calibration, we introduce the CalibratedMath suite of tasks. We compare the calibration of uncertainty expressed in words (\"verbalized probability\") to uncertainty extracted from model logits. Both kinds of uncertainty are capable of generalizing calibration under distribution shift. We also provide evidence that GPT‑3's ability to generalize calibration depends on pre-trained latent representations that..."},"evidence_pages":[{"url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","final_url":"https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words","title":"Teaching models to express their uncertainty in words","http_status":200,"content_type":null,"capture_method":"exa","fetched_at":"2026-06-08T15:46:48.454+00:00","bytes":null,"raw_path":null,"content_hash":null,"excerpt_chars":1200,"truncated":true,"excerpt":"Teaching models to express their uncertainty in words | OpenAI May 28, 2022 Teaching models to express their uncertainty in words Loading… Share Abstract We show that a GPT‑3 model can learn to express uncertainty about its own answers in natural language—without use of model logits. When given a question, the model generates both an answer and a level of confidence (e.g. \"90% confidence\" or \"high confidence\"). These levels map to probabilities that are well calibrated. The model also remains moderately calibrated under distribution shift, and is sensitive to uncertainty in its own answers, rather than imitating human examples. To our knowledge, this is the first time a model has been shown to express calibrated uncertainty about its own answers in natural language. For testing calibration, we introduce the CalibratedMath suite of tasks. We compare the calibration of uncertainty expressed in words (\"verbalized probability\") to uncertainty extracted from model logits. Both kinds of uncertainty are capable of generalizing calibration under distribution shift. We also provide evidence that GPT‑3's ability to generalize calibration depends on pre-trained latent representations that..."}],"related_signals":[{"id":"b3668d3b-26d2-40c0-9d4f-ed1a67927aa4","url":"https://onlylabs.fyi/signals/b3668d3b-26d2-40c0-9d4f-ed1a67927aa4","source_url":"https://openai.com/index/supporting-eu-trustworthy-ai-ecosystem","title":"Supporting Europe’s work in ensuring a trustworthy AI ecosystem ","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-11T00:00:00+00:00","first_seen_at":"2026-06-11T08:00:56.140796+00:00","date_source":"rss.item_date"},{"id":"2638c0a7-b372-409c-ac72-f6d81d6464dc","url":"https://onlylabs.fyi/signals/2638c0a7-b372-409c-ac72-f6d81d6464dc","source_url":"https://openai.com/index/using-codex-to-simulate-black-holes","title":"How an astrophysicist uses Codex to help simulate black holes","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-11T00:00:00+00:00","first_seen_at":"2026-06-11T07:01:16.936464+00:00","date_source":"rss.item_date"},{"id":"509ea784-51ec-4ede-855b-5a4d1b27d3be","url":"https://onlylabs.fyi/signals/509ea784-51ec-4ede-855b-5a4d1b27d3be","source_url":"https://openai.com/index/openai-on-oracle-cloud","title":"Access OpenAI models and Codex through your Oracle cloud commitment","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-10T20:00:00+00:00","first_seen_at":"2026-06-11T07:01:16.936464+00:00","date_source":"rss.item_date"},{"id":"4f051449-87f2-466e-941e-b5918381a8fe","url":"https://onlylabs.fyi/signals/4f051449-87f2-466e-941e-b5918381a8fe","source_url":"https://openai.com/index/prc-linked-influence-operations-ai-debates","title":"PRC-linked influence operations are targeting AI debates in the US","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-10T12:00:00+00:00","first_seen_at":"2026-06-11T07:01:16.936464+00:00","date_source":"rss.item_date"},{"id":"4507c0c1-cb74-4bb3-b62b-5f6c2d37e20d","url":"https://onlylabs.fyi/signals/4507c0c1-cb74-4bb3-b62b-5f6c2d37e20d","source_url":"https://openai.com/index/lseg","title":"From data to decisions: how LSEG is scaling trusted AI","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-10T00:00:00+00:00","first_seen_at":"2026-06-10T09:18:54.26094+00:00","date_source":"rss.item_date"},{"id":"fb16aa7a-c4ef-4859-b514-0839c2f1330d","url":"https://onlylabs.fyi/signals/fb16aa7a-c4ef-4859-b514-0839c2f1330d","source_url":"https://openai.com/index/nextdoor","title":"How engineers at Nextdoor use Codex to build without limits","context":null,"kind":{"key":"post_published","label":"Writing"},"org":{"slug":"openai","name":"OpenAI","category":"frontier-lab"},"occurred_at":"2026-06-09T12:00:00+00:00","first_seen_at":"2026-06-10T07:01:28.700378+00:00","date_source":"rss.item_date"}]}