Model size is not the primary factor in outcome-prediction performance¶

Claim · All claims
Evidence · 1 study · 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? The larger Llama-3.3-70B underperformed the smaller Llama-3.2-3B on outcome prediction, suggesting architectural choices and training refinements matter more than scale alone. → Michal Štefánik 2026

Evidence¶

Michal Štefánik 2026¶

Michal Štefánik, Jan Nehyba, Jirina Karasova, Martin Fico, Lucie Škarková, Markéta Košatková, David Kosatka. (2026). Edustories: A Collection of Real-world Case Studies from Classroom Practices. https://arxiv.org/abs/2609.20484 doi:10.48550/arxiv.2609.20484

q2 · i? · design · r2

Comparative analysis of the six evaluated models' accuracies on the 310-case benchmark. The article reports the size inversion as an observed result and hypothesizes that "outcome prediction may depend more on general reasoning capability".

"For example, the larger Llama-3.3-70B underperforms the smaller Llama-3.2-3B, suggesting that architectural choices, training data and other training refinements play a more important role than scale alone."

Discussion¶

Related Claims¶

• Current LLMs fall short of human experts in predicting classroom intervention outcomes — related
• Under identical LoRA hyperparameters and training data, the 27B Gemma model outperforms the 70B LLaMA model on every computed essay-scoring metric — related
• LLaMA's LoRA adaptation compresses its prediction range to roughly 2.0–4.0, while Gemma maintains a broader distribution closer to the ground truth — related
• CDPK performance scales with model size, with a sharp Pareto-frontier drop-off below around 8B parameters — reports the opposite
• Larger Llama Guard models outperform smaller ones on education-prompt classification, with the 8B model best in accuracy, recall, and F1 — reports the opposite
• Among the three 8B LLMs, Llama 3 underperforms Apertus and Granite 3.3 on most evaluation metrics — related
• LoRA fine-tuning hyperparameters are not model-agnostic: increasing rank above 64 destroyed LLaMA's feedback generation while Gemma remained robust — related