37% on ARC-AGI-2 Kaggle Without LLMs
Web page · https://twitter.com/evilmathkid/status/2120581574584582301
First seen by Oriel on .
AI summary
Mithil Vakde reports scoring 37% on ARC-AGI-2 Kaggle with a 100M-parameter looped transformer trained from scratch without LLMs.
- The entry peaked at sixth place and reportedly exceeded the previous year's winning LLM.
- Adding looping during the final week raised the score from 33% to 35%, then 37%.
- Vakde describes the result as likely the competition's highest score achieved without an LLM.
- The linked blog post provides a writeup for the paper track.
#arc-agi-2 #transformers #machine learning #kaggle
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 3, latest observation ; 0 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 0, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50461495 | https://twitter.com/evilmathkid/status/2120581574584582301 Read discussion |