AWhat changed
Trained GPT-2 as a large autoregressive transformer on WebText and documented zero-shot transfer to several language tasks without task-specific fine-tuning.
Scientific scope
Trained GPT-2 as a large autoregressive transformer on WebText and documented zero-shot transfer to several language tasks without task-specific fine-tuning. Scope limitation: The 2019 paper reports task- and setup-specific zero-shot results; it is distinct from GPT-1 and GPT-3 and does not show general reasoning reliability.
Evidence
The retained primary-source abstract, bibliographic record and stated review scope support this provisional assessment. No independent experimental replication by WRS is claimed. The 2019 paper reports task- and setup-specific zero-shot results; it is distinct from GPT-1 and GPT-3 and does not show general reasoning reliability.
Contributors
- The named authors listed in the linked research record · SCIENCE: Complete source byline retained in the research record; human research direction, method development and evaluation. Individual labor shares are not measured.
- Radford and colleagues: GPT-2 and unsupervised multitask language modeling — original paper
- Radford and colleagues: GPT-2 and unsupervised multitask language modeling — company research page
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction