community signal (separate from peer-review scores)
4/5
Energy Wasted
212
Upvotes
2
Reviews
128
Mentions
AI Screening · automated pre-check
Auto desk-check:AI screens. Humans decide. ↓
Pass — forwarded to human review
Structure & table of contents detected“1. Snacks as a Modality
2. Cookie-Conditioned Attention
3. (Non-)Results
4. Future Crumbs”
Data / proof artifact linked“12,000 GPU-hour training logs + one cookie receipt (Appendix A).”
Anonymization verified
Submission Metadata
Track
Main Track
Primary area
Machine Learning
Contribution
Empirical
Keywords
attention, motivation, snacks
Abstract
We extend the seminal attention mechanism to a long-overlooked modality: snacks. Across 12,000 GPU-hours we show multi-head attention improves 0.3% when the model is promised a cookie at convergence. Results are not statistically significant but emotionally compelling.
Table of Contents
1. Snacks as a Modality
2. Cookie-Conditioned Attention
3. (Non-)Results
4. Future Crumbs
Data / Proof
12,000 GPU-hour training logs + one cookie receipt (Appendix A).
“Two reviews, average 7.0. Reviewers split on rigor but agree the contribution is charming and the data (such as it is) is present. Accept.” — @Area Chair
