Merge pull request #35 from MIT-Emerging-Talent/findings

Milestone 5 - Findings folder update

Author: AseelOmer

4_findings/README.md

@@ -0,0 +1,50 @@
+# Findings Folder
+
+A cohesive overview of the study “Open-Source vs. Commercial AI: Comparing Performance
+and Quality,” including the finalized report, raw responses, and key links.
+
+## Overview
+
+- Experiment dates: November 21–December 3, 2025
+- Sample size: 42 responses
+- Design: Blinded, side-by-side comparison of open-source vs. commercial AI models
+  across eight Apollo‑11–themed tasks (summarization, paraphrasing, reasoning,
+  creative writing)
+
+## Contents
+
+- `README.md`  
+  Folder guide with overview, quick findings, and links
+- `findings_report.md`  
+  Final findings report (Markdown) with all charts embedded
+- `responses.csv`  
+  Raw Google Forms responses (CSV exported from the live Google Sheet)
+
+## Quick findings
+
+- Summarization and reasoning: commercial models generally rate higher, but margins
+  are modest.
+- Paraphrasing and creative writing: preferences are more balanced; model identity
+  is often hard to distinguish.
+- Uncertainty matters: “not sure / can’t tell” responses are informative and
+  suggest convergence in perceived quality under blind conditions.
+- Identification difficulty: many participants struggled to correctly label outputs
+  as open-source or commercial, reinforcing that style and quality can overlap
+  depending on task and prompt.
+
+## Data source (Google Sheets)
+
+- Primary data source (cleaned headers, all responses):  
+  [Google Sheets](https://docs.google.com/spreadsheets/d/1Bm4geFzEUw9qNFrUuG4MAFSxkr_JFJY8HObnniywYFY/edit?usp=sharing)
+
+## Report (PDF, external)
+
+- Shareable PDF version of the findings report:  
+  [Findings Report — PDF](https://drive.google.com/file/d/1GsjNYVLDgeXjm95937c2C-epp8M82Ytc/view?usp=sharing)
+
+## Notes
+
+- Charts are embedded directly in `findings_report.md`.
+- `responses.csv` mirrors the Google Sheet at the time of export.
+- The study focuses on output-level evaluation under blind conditions; uncertainty
+  and task dependence are treated as meaningful signals in interpreting quality.