On benchmarks, GPT-3.5-turbo outperformed the fine-tuned Phi model in knowledge and story completion, while PhinetunEd scored best in story generation¶
Claim · All claims
Evidence · 1 study · designr2·q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person · recheck by 2028: tool and model accuracy changes quickly
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q2 i? The GPT model performed better in knowledge (MMLU 70% vs 62% base, 60% PhinetunEd) and story completion (Hellaswag 85.5% vs 59%/58%), whereas the fine-tuned model scored best in story generation (7.46 vs 7.05 and 6.87). → Tozadore 2025
Evidence¶
Tozadore 2025¶
Tozadore, D., Ertug, N., Chaker, Y., & Abderrahim, M. (2025). RoboBuddy in the Classroom: Exploring LLM-Powered Social Robots for Storytelling in Learning and Integration Activities. https://arxiv.org/abs/2508.16706 doi:10.48550/arxiv.2508.16706
q2 · i? · design · r2
Model evaluation reported in Table I: MMLU, Hellaswag, and GPT-4-turbo-judged story generation on 200 WritingPrompts prompts rated 1–10. Printed scores: GPT-3.5-turbo 70%/85.5%/6.87; Phi-3 base 62%/59%/7.05; PhinetunEd 60%/58%/7.46. The comparison names GPT-3.5-turbo and PhinetunEd as the two sides.
"The benchmark evaluation showed that the GPT model performs better in knowledge, reasoning, and story completion, whereas our finetuned models score best in story generation."
Discussion¶
Related Claims¶
- A LoRA-adapted open-weight Gemma-3-27B model fine-tuned on 120 essays outperforms a fine-tuned GPT-3.5 baseline on TOEFL essay scoring across all reported metrics — related
- The OpenAI gpt-4o reference model outperformed all quantised models on human-rated coherence, fluency, relevance and cultural appropriateness, serving as an upper benchmark — related
- On question answering and commonsense reasoning the model outperforms previous best results by up to 8.9% on Story Cloze and 5.7% overall on RACE — related
- No significant differences emerged between GPT-3.5 Turbo and Claude 3 Opus on affective, collaborative, or performance measures — related
- Fine-tuned 8B LLMs generate K-2 stories with difficulty-related metric scores statistically significantly different from (mostly better than) zero-shot GPT-4o and Llama 3.3 70B — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
phinetuned-benchmark-comparison-gpt |
title |
On benchmarks, GPT-3.5-turbo outperformed the fine-tuned Phi model in knowledge and story completion, while PhinetunEd scored best in story generation |
description |
On benchmarks, GPT-3.5-turbo outperformed the fine-tuned Phi model in knowledge and story completion, while PhinetunEd scored best in story generation |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2025 |
recheck_by |
2028 |
generated |
by process:wiki-ingest on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — tozadore-2025 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.