GPT-4 outperforms GPT-3.5-turbo in a single-lesson question-detection test¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? In a one-lesson comparison, GPT-4 five-shot CoT achieved higher precision, recall, balanced accuracy and F1 than GPT-3.5-turbo five-shot CoT. → Lehesvuori 2026

Evidence¶

Lehesvuori 2026¶

Lehesvuori, S., Urrutia, F., Heilala, V., Araya, R., & Hämäläinen, R. (2026). Discovering technology-aided possibilities for automatic analysis of science teacher questions. Education Tech Research Dev. https://doi.org/10.1007/s11423-025-10545-3

q2 · i? · design · r2

Single-lesson test (Table 4): GPT-4 5-shot CoT reached F1 0.5797 and BA 0.9145 versus F1 0.51 for GPT-3.5-turbo, showing "GPT-4 outperforming the previous model."

"The one lesson results show GPT-4 outperforming the previous model. Results also indicate that GPT-4 could challenge the HCR with the whole data, yet this experiment cannot be executed here since the high costs"

Discussion¶

Related Claims¶

• In detecting teacher questions with zero-shot GPT-3.5-turbo, chain-of-thought prompting had slightly higher F1 and precision than Self-Ask but slightly lower recall and balanced accuracy — related
• Few-shot chain-of-thought prompting outperforms zero-shot prompting for detecting teacher questions — related
• Few-shot LLM approaches outperform standalone supervised ML approaches, yet hand-crafted rules remain relevant — related
• Curricular chain-of-thought prompting improves competency-classification accuracy over zero-shot, with gains concentrated in larger models, while definition-based prompting does not improve performance — related