UCLA and AWS paper: AI agents find math shortcuts but thinking time often cancels the gains
rohanpaul_ai · x · 2026-11-17
A new UCLA and AWS paper, Agentic Computational Algebra, frames agent evaluation around time, not correctness: general algorithms are guaranteed but slow, while reformulations exploiting hidden structure can be much faster.
They gave Claude Opus 5 agents polynomial systems with hidden shortcuts, counting thinking time as cost, across three modes: fresh solving, solving related instances with persistent memory, and compiling a reusable solver upfront.
Key findings:
- The agent solved 103 of 108 systems within 600s vs 60 for plain msolve, but was faster on only 12 of those 60 — reasoning overhead often cancels the shortcut's savings.
- Memory reduces repeated solving cost but can preserve suboptimal strategies.
- Compiled solvers ran 5.4x faster than a baseline given the known shortcut, after 11.4 hours of setup.
Practical takeaway: use agents on problems standard solvers can't finish, and build a reusable solver only when enough similar problems repay the setup cost.
More from coding & agent
- Ripple-65B Hackathon: Self-Quantization Shrinks Model from 60GB to 30GB with Little Quality Loss — Azaliamirh · 2026-11-17
- Treating agent orchestration like a factory: the 7 jobs of an orchestration layer — blaizedsouza · 2026-11-17
- Pixel art + Gaussian splats game experiments: all code, no assets, but roadblocks ahead — algo_diver · 2026-11-17
- Fable 5.2 praised as first model that ignores unimportant feedback and tells you to ship — neuroecology · 2026-11-17
- OpenAI's jxnlco on building taste: 'If you want good taste, you have to eat' — jxnlco · 2026-11-17
- I ran 3,360 experiments to see if AI agents recommend the right products — santiviquez · 2026-11-17