Public description
Company overview
Snorkel AI builds specialized training data, benchmarks, and evaluation environments that help frontier models and agents perform in high-stakes domains. Evaluating AI agents on long-horizon, economically valuable professional workflows with verifiable outcomes. Evaluating whether coding agents can plan, execute, verify, and recover across complex terminal-native engineering tasks. Consistent reward hacking (e.g. copying fixes from newer PyPI/npm releases) detected, 5 tasks removed from score Opus 5.5 39.3% 2 GPT-6 Astra 36.8% 3 Fable 5.1 34.7% 4 Fable 5 34.7% 5 Opus 5 34.7% Knowledge Work WorkplaceAgents Evaluates AI agents on economically valuable professional work and verifiable workplace deliverables. Evaluates agents on building financial models in spreadsheets, across 101 expert-written tasks. of The Frontier AI Data Lab Data development for the frontier Snorkel partners with frontier AI teams to build research-grade datasets, evaluation systems, and runnable environments where generic coverage runs out. Website: https://snorkel.ai/ Source: Company site
Company
Company overview
Snorkel started with a contrarian bet at the Stanford AI Lab: that training data, not models or compute, would decide whether machine learning worked. The 2017 VLDB paper that introduced Snorkel made the bet a system: data programming, weak supervision, and the radical idea that you could build state-of-the-art models without hand-labeling a single example. A decade and 250+ publications later and awards at NeurIPS, ICML, ICLR, UAI, VLDB, and more, the bet has only gotten stronger as models have gotten bigger.
Snorkel
Company · Subject
| Schema | Name | Relation |
|---|---|---|
| Company | Snorkel | Subject |
