Why Can I Easily Find Code Optimizations That Frontier LLMs Overlook?
oyacaro · x · 2026-09-17
A developer describes a recurring pattern: LLMs fail at his optimization tasks, yet when he suggests a direction, the model enthusiastically agrees — and the idea genuinely works and performs much better. Despite not being a top engineer, he routinely spots improvements frontier models miss, and wonders what exactly he brings to the table. The post touches on sycophancy and the human's real value in AI-assisted coding.
More from coding & agent
- From babysitting simple features to full monoliths by agents in 8 months — reach_vb · 2026-09-17
- Antigravity 0-day shows how DNS rebinding on loopback turns AI coding agents into RCE — evilsocket · 2026-09-17
- Paper2Agent published in Nature: turning papers into collaborating virtual-author agents — james_y_zou · 2026-09-17
- Astra designs, Luna executes: a quota-saving multi-agent coding workflow with scheduled reviews — TheMoonMidas · 2026-09-17
- Dev Open-Sources Blue to Fix MCP Delivery and Config Pain in Inference Gateways — Educational-Hold9425 · 2026-09-17
- RL-Only Post-Training Lifts Kimi K2.7 Past GPT-5.6 and Kimi K3 on Coding Benchmarks — echen · 2026-09-17