All Posts

BITCOS

BITCOS and the Case for Going Below 1.58 Bits

Ternary weights do not always need 1.58 bits of storage each. BITCOS explores preserving those weights while shrinking …

BITCOS LLM Compression Ternary Quantization
NVIDIA

Rust Could Make CUDA Harder to Leave

Writing GPU code in Rust can make CUDA more approachable while leaving switching costs intact. Hardware flexibility …

NVIDIA CUDA Rust
PostgreSQL

QORL 4B and the Fine Print Behind '81% Faster'

QORL 4B’s claim of 81% faster query plans needs a precise definition. Fair benchmarks, performance regressions, and …

PostgreSQL AI Query Optimization
Mistral

Browser AI Has to Earn the Right to Read

What would make a hypothetical Mistral–Mozilla browser AI trustworthy? Start with what it can read, where it sends data, …

Mistral Mozilla Browser AI
AI Security

The AI Hacked It. Now Show the Setup.

AI hacking claims depend on starting permissions, human assistance, and what actually succeeded. Evaluators and model …

AI Security Irregular AI Safety Evaluations