Overall quality was similar across durations and across the 21-lecture held-out set unused during system development¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Medium (5–15 minute) and long videos scored similarly on global quality (3.36 vs. 3.53; Mann–Whitney U p = 0.60), and videos from the 21 held-out lectures scored G = 3.36 versus 3.51 on the 10 development lectures. → Romain Puech 2026

Evidence¶

Romain Puech 2026¶

Romain Puech, Dewang Kumar Agarwal, Antonio Santamaría Escobar, Dimitris Bertsimas. (2026). Bespoke: Generating MOOC-Quality Industry-Personalized Lecture Videos at Scale. arXiv preprint. https://arxiv.org/abs/2609.26540 doi:10.48550/arxiv.2609.26540

q2 · i? · design · r2

Between-group comparisons within the 92-video expert rating corpus, reported with Mann–Whitney U tests and confidence intervals. The article reports the medium-versus-long difference of 0.17 with p = 0.60, and separately the held-out-versus-development difference of −0.15 (95% CI[−0.48, 0.18]).

"Medium and long videos are similarly close (3.36 vs. 3.53; Mann–Whitney𝑈 𝑝= 0.60; difference0.17, 95% CI[−0.17, 0.50]; Table 2)."

Romain Puech 2026 (2)¶

Romain Puech, Dewang Kumar Agarwal, Antonio Santamaría Escobar, Dimitris Bertsimas. (2026). Bespoke: Generating MOOC-Quality Industry-Personalized Lecture Videos at Scale. arXiv preprint. https://arxiv.org/abs/2609.26540

q2 · i? · design · r2

Generalizability check comparing the 21 lectures unused while developing the pipeline against the 10 development lectures, using the same expert ratings. The article reports G = 3.36 versus 3.51, a difference of −0.15, and concludes overall quality does not drop sharply on new lectures.

"Videos from the 21 held-out lectures score 𝐺= 3.36, versus 3.51on the 10 development lectures (difference −0.15, 95% CI [−0.48, 0.18])."

Discussion¶

Related Claims¶

• Domain-matched experts judged 87% of Bespoke-generated lecture videos at or above the rubric midpoint corresponding to a standard MOOC lecture's quality (mean G = 3.42 of 5) — related
• Students reported that long recorded lectures impeded focus, and one student appreciated an instructor's choice to break lectures into shorter videos — related
• MOOC video engagement declines with video length, with median engagement under half the duration for videos longer than nine minutes — related