A vibe-research experiment: one pairwise comparison study shipped on ~1.5 hours of attention
maxsloef · x · 2026-09-27
maxsloef shares a small "vibe-research" experiment, deliberately applying a 90/10 approach: ship when done, not when perfect. He estimates the whole project took about 1.5 hours of total attention spread across three days.
The post links to the full writeup, the pairs and ratings, and the code to run it yourself—demonstrating a reproducible workflow for shipping tiny AI-assisted research projects at minimal cost.
Related event: Blogger runs Namedrop eval in 1.5 hours of focused attention(4 posts)→
More from Research
- Interpretability researcher: sandbagging signals from probes would block model deployment — thebasepoint · 2026-09-27
- Andrew Dai: pre-training + fine-tuning was born from a forgotten checkpoint bug — ziv_ravid · 2026-09-27
- AI co-scientists can hypothesize, design experiments and analyze data, but humans still decide what makes sense, Nature reports — irinarish · 2026-09-27
- Viral claim: Google's ScientistTwo runs end-to-end autonomous ML research with no humans — mtizard · 2026-09-27
- ScienceArena benchmark: LLMs score 64.5% on chemistry tasks needing structural diagrams vs 74.1% without — geoffwolfe · 2026-09-27
- Experiment: a learned memory table transfers across model sizes with a small adapter — sn2006gy · 2026-09-27