-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathpromptedu.html
More file actions
147 lines (132 loc) · 10.9 KB
/
Copy pathpromptedu.html
File metadata and controls
147 lines (132 loc) · 10.9 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>Ricieri, de Souza et al. (2024) — PROMPT-EDU — Research Brief</title>
<style>
@page { size: Letter; margin: 0.5in; }
:root { --ink: #0a0a0a; --muted: #444; --rule: #888; --accent: #1f3a5f; --bg-panel: #f3f3f0; }
* { box-sizing: border-box; }
html, body { margin: 0; padding: 0; font-family: "Charter", "Iowan Old Style", "Palatino Linotype", Georgia, "Times New Roman", serif; color: var(--ink); font-size: 10.5pt; line-height: 1.32; -webkit-print-color-adjust: exact; print-color-adjust: exact; }
body { padding: 0.4in; max-width: 8.5in; }
h1 { font-size: 16pt; margin: 0 0 2pt 0; line-height: 1.18; font-weight: 700; }
h1 .sub { display: block; font-size: 11pt; font-weight: 400; color: var(--muted); font-style: italic; margin-top: 2pt; }
.meta { font-size: 9.5pt; color: var(--muted); margin-bottom: 8pt; padding-bottom: 6pt; border-bottom: 0.5pt solid var(--rule); }
.meta strong { color: var(--ink); }
h2 { font-size: 10pt; text-transform: uppercase; letter-spacing: 0.06em; color: var(--accent); border-bottom: 0.5pt solid var(--accent); padding-bottom: 1pt; margin: 10pt 0 4pt 0; font-weight: 700; }
p { margin: 0 0 5pt 0; }
ul { margin: 0 0 5pt 0; padding-left: 14pt; }
li { margin-bottom: 2pt; }
.grid { display: grid; grid-template-columns: 1fr 1fr; gap: 12pt; }
.panel { background: var(--bg-panel); border-left: 2pt solid var(--accent); padding: 6pt 8pt; font-size: 9.8pt; }
.panel h2 { margin-top: 0; border-bottom: none; padding-bottom: 0; }
.keywords { font-size: 9.2pt; color: var(--muted); margin-top: 4pt; }
.keywords code { display: inline-block; padding: 1pt 5pt; background: #ececea; border-radius: 3pt; margin: 1pt 2pt 1pt 0; font-family: "SF Mono", Menlo, Consolas, monospace; font-size: 8.5pt; color: var(--ink); }
.findings { counter-reset: f; list-style: none; padding-left: 0; }
.findings li { counter-increment: f; padding-left: 18pt; position: relative; margin-bottom: 3pt; }
.findings li::before { content: "F" counter(f); position: absolute; left: 0; top: 0; font-family: "SF Mono", Menlo, monospace; font-size: 8pt; color: var(--accent); font-weight: 700; }
.footer { margin-top: 10pt; padding-top: 5pt; border-top: 0.5pt solid var(--rule); font-size: 8.5pt; color: var(--muted); display: flex; justify-content: space-between; }
.no-print { display: none; }
@media screen {
body { background: #e9e8e4; }
.page { max-width: 8.5in; min-height: 11in; margin: 16px auto; padding: 0.5in; background: white; box-shadow: 0 2px 14px rgba(0,0,0,0.12); }
.no-print { display: block; max-width: 8.5in; margin: 16px auto 0; padding: 10px 14px; background: #fff8d6; border: 1px solid #e0c441; font-family: -apple-system, system-ui, sans-serif; font-size: 13px; line-height: 1.45; border-radius: 4px; }
.no-print kbd { background: #fff; border: 1px solid #999; border-radius: 3px; padding: 1px 5px; font-family: monospace; font-size: 12px; }
}
</style>
</head>
<body>
<div class="no-print">
<strong>To save as PDF:</strong> press <kbd>Cmd</kbd>+<kbd>P</kbd>, choose <em>"Save as PDF"</em>. This yellow box won't appear in the PDF.
| <a href="promptedu-pt.html" style="font-weight:600;">🇧🇷 Versão em Português</a>
</div>
<div class="page">
<h1>
PROMPT-EDU: Fine-Tuning Command Script for ChatGPT
<span class="sub">A Structured Prompt Sequence for Optimizing ChatGPT Fine-Tuning in Educational Contexts</span>
</h1>
<div class="meta">
<strong>Denise da Vinha Ricieri</strong> (first author),
Adriana M. G. de Farias, <strong>Fabiano Rodrigues de Souza</strong>, Raphaela V. G. Barreto
·
<strong>SCIAS Educação, Comunicação e Tecnologia</strong> — v.6, n.1, p.107–138, jan./jun. 2024
·
DOI: <a href="https://doi.org/10.36704/sciaseducomtec.v6i1.8374" target="_blank">10.36704/sciaseducomtec.v6i1.8374</a>
·
Peer-reviewed journal article · e-ISSN: 2674-905X
</div>
<h2>Research Question</h2>
<p>
Can a structured, sequenced set of prompts — the Prompt-EDU Script — reliably optimize the fine-tuning (FT) of ChatGPT for educational contexts, including for novice users with accounts that have minimal prior FT history?
</p>
<h2>Methodology</h2>
<ul>
<li><strong>Design:</strong> Qualitative descriptive experimental study tested across multiple ChatGPT versions in Brazil and the USA.</li>
<li><strong>Prompt types used:</strong> In-Context Learning (ICL), Chain of Thought (CoT), and Demo prompts combined in a structured sequence.</li>
<li><strong>Three-profile Prompt-EDU Script:</strong> Context prompt → Demo-CoT prompts → Educational prompt (using Bloom's Taxonomy verbs).</li>
<li><strong>Test scenario:</strong> Simulated dialogue between Paulo Freire and Mark Zuckerberg on AI and Education — same scenario as CoBICET 2023 paper, now used to validate the script's FT performance.</li>
<li><strong>Bias addressed:</strong> Non-native language processing bias in GPT (double translation pipeline: input language → EN → output language). Bloom's Taxonomy verbs used to override ambiguous verb interpretation.</li>
<li><strong>Accounts tested:</strong> IAR1-BRA (Brazil, GPT-3.5) · IAR2-BRA (Brazil, GPT-4) · IAT-EUA (USA, GPT-4) — beginner FT accounts.</li>
</ul>
<div class="grid">
<div>
<h2>Key Findings</h2>
<ul class="findings">
<li>Prompt-EDU Script <strong>succeeded on all established analytical markers</strong> across all tested ChatGPT accounts and versions.</li>
<li>Performance matched <strong>deep machine learning benchmarks</strong> described in prior literature for structured prompt sequences.</li>
<li>A single Demo prompt + well-structured CoT prompt <strong>outperformed multi-demo ICL</strong> — replicating Chen et al. (2023) findings.</li>
<li><strong>Training bias detected and corrected:</strong> Brazil account (IAR1-BRA) initially used GPT-3.5 mid-FT — bias isolated and script adapted.</li>
<li><strong>Bloom's Taxonomy verbs</strong> effectively controlled AI interpretation of educational verbs across the non-native language (Portuguese) barrier.</li>
<li>Script proven <strong>reliable for novice users</strong> optimizing FT for teaching-learning themes in Portuguese-speaking contexts.</li>
</ul>
</div>
<div class="panel">
<h2>The Prompt-EDU Script (3 Profiles)</h2>
<p><strong>Profile 1 — Context Prompt:</strong> Establishes the educational domain, the task, and the AI persona as an educational assistant. Sets the FT direction.</p>
<p><strong>Profile 2 — Demo-CoT Prompt:</strong> Provides a Chain of Thought demonstration that trains the model to reason in educational terms before producing output.</p>
<p><strong>Profile 3 — Educational Prompt:</strong> Uses Bloom's Taxonomy action verbs to assign a specific cognitive-level task, overriding language ambiguity and producing educationally valid output.</p>
<p style="margin-top:5pt;font-size:9pt;color:var(--muted);">Tested: ChatGPT 3.5 & 4 · Brazil & USA accounts · Beginner FT level</p>
</div>
</div>
<h2>Technical Concepts & Skills</h2>
<div class="keywords">
<strong style="font-size:8.5pt;color:var(--accent)">AI & Machine Learning:</strong>
<code>LLM</code> <code>ChatGPT</code> <code>Fine-Tuning (FT)</code> <code>Deep Learning</code> <code>Transformer Architecture</code> <code>NLP</code> <code>Neural Networks</code> <code>Generalization</code>
<br><strong style="font-size:8.5pt;color:var(--accent)">Prompt Engineering:</strong>
<code>In-Context Learning (ICL)</code> <code>Chain of Thought (CoT)</code> <code>Demo Prompts</code> <code>Prompt-EDU Framework</code> <code>Educational Prompts</code> <code>Prompt Sequencing</code>
<br><strong style="font-size:8.5pt;color:var(--accent)">Education & Pedagogy:</strong>
<code>Bloom's Taxonomy</code> <code>Instructional Design</code> <code>AI in Education</code> <code>Non-Native Language Bias</code> <code>Teacher Training</code> <code>Systematization</code>
</div>
<h2>Why This Matters for Educators</h2>
<p>Most educators using ChatGPT for lesson planning, content creation, or student feedback are working with accounts that have minimal fine-tuning — and they are writing prompts in non-English languages, which introduces a hidden double-translation bias. The Prompt-EDU Script is the first peer-reviewed, tested methodology specifically designed to overcome both problems: it gives educators a reliable, systematic way to train ChatGPT to behave as a genuine educational assistant, using the international language of pedagogy (Bloom's Taxonomy) to bridge the language gap.</p>
<h2>Selected References</h2>
<p style="font-size:9pt;color:var(--muted);">
Chen et al. (2023) — Many-shot vs. single Demo ICL performance ·
Wei et al. (2022) — Chain of Thought prompting ·
Liu et al. (2023) — Pre-train, prompt, and predict (ACM Comp. Surv.) ·
Wang et al. (2023) — LLMs as implicit topic models ·
Vaswani et al. (2017) — Attention is all you need ·
Liang et al. (2023) — GPT detectors biased against non-native writers ·
Armstrong (2010) — Bloom's Taxonomy revised ·
Kasneci et al. (2023) — ChatGPT for good? (Learning & Individual Differences) ·
OpenAI Inc. (2023) — GPT-4 Technical Report.
</p>
<h2>Read the Full Paper</h2>
<p>
<strong>ResearchGate (open access):</strong>
<a href="https://www.researchgate.net/publication/382859526" target="_blank">researchgate.net/publication/382859526</a><br>
<strong>Journal:</strong> SCIAS Educação, Comunicação e Tecnologia — v.6, n.1, 2024 — DOI: <a href="https://doi.org/10.36704/sciaseducomtec.v6i1.8374" target="_blank">10.36704/sciaseducomtec.v6i1.8374</a>
</p>
<h2>Author Bio</h2>
<p><strong>Fabiano Rodrigues de Souza</strong> — PhD in Biotechnology · Harvard Graduate School of Education · Dual MBAs · Kirkpatrick Certified Professional · 2× Braskem Award Winner · Editorial Board Member since 2006. Researcher at the intersection of AI, neuroscience, and multilingual education.</p>
<div style="margin-top:8pt;padding:5pt 8pt;background:#f3f3f0;border-left:2pt solid #1f3a5f;font-size:9pt;">
<strong>See also — Paper 1:</strong> <a href="index.html">Simulating Dialogues and Characters in ChatGPT-4 (CoBICET 2023)</a> — multilingual GPT-4 evaluation: English vs. Portuguese output quality, double-translation bias, NLP bias taxonomy.<br>
<strong>See also — Paper 2:</strong> <a href="emo-ai.html">EMO-AI Teaching Loop (UNICAMP X Inovações Curriculares 2025)</a> — neurodidactic microlearning via WhatsApp & GenAI: 21% enthusiasm, 19% curiosity, correlation = 1.0.
</div>
<div class="footer">
<span>Ricieri, de Souza et al. (2024) — PROMPT-EDU — SCIAS Educação, Comunicação e Tecnologia</span>
<span><a href="promptedu-pt.html">Versão em Português</a></span>
</div>
</div>
</body>
</html>