Kimi K2.7-Code cuts thinking tokens 30% — but practitioners say the benchmarks don't check out

Kimi K2.7-Code claims 30% fewer thinking tokens and a drop-in API swap path, but independent benchmarks show kernel ...

Microsoft’s open-source SkillOpt automatically upgrades AI agent skills without touching model weights

Microsoft's SkillOpt brings deep-learning discipline to AI agent skills, replacing manual prompt tweaking with mathematically ...
IEEE

Centroiding Error Modeling and Code Optimization for Restored Star Images Using Coded Exposure in High-Dynamic Conditions

Abstract: Under high-dynamic conditions, star sensors often capture images with low signal-to-noise ratio (SNR). Image restoration can mitigate energy dispersion from star trailing, but it is ...