Talk:Knowledge cutoff/GA1

Article (edit | visual edit | history) · Article talk (edit | history) · Watch

Nominator: 16dvnk (talk · contribs) 03:58, 16 July 2025 (UTC)

Reviewer: David Eppstein (talk · contribs) 00:18, 20 July 2025 (UTC)

There are multiple unsourced statements. In some cases this is visible from a lack of footnote; in other cases, the next footnote covers only a later unrelated claim

• One sentence at end of section "Factors behind knowledge cutoffs"
• Two sentences at end of section "Knowledge gaps"
• One sentence at start of section "Historical context"

Twelve of 16 references appear not to meet our standards for reliable sources

• [1] Conductor, [8] ProjectPro (commercial sales site)
• [2-4, 7, 9-11] OpenAI, Anthropic, Google AI, Amazon primary sources about their own products
• [5] Otterly, blog
• [12] WP:FORBES
• [16] non-peer-reviewed preprint

Reference [6] (Brown et al.) appears reliable but cannot be used for the claim that it originated the idea of a knowledge cutoff (in the infobox); we need independent secondary sourcing for that.

Spot-checking the next use of [6], for the claims "Training large language models on static datasets is standard practice. This is necessary for achieving reproducibility and stability in performance evaluation." found no use of the words static and reproducible, and the only uses of the word stable referring to the hardware platform and not performance evaluation.

Reference [6] is also used for the claim "The practice of announcing a cutoff date became an industry standard for transparency after the release of GPT-3 in 2020." as a 2020 publication it seems an unlikely choice for a source about what became standard after 2020 and the word announce does not appear in it.

I conclude that this is very far from Good Article criterion 2 (sourcing) and falls under WP:GAFAIL #1. It was not ready for a Good Article nomination. —David Eppstein (talk) 00:25, 20 July 2025 (UTC)