Wen-Ting Zhao

Personalized memory gating unlocks more relevant long-term recall by learning from individual user interaction patterns, outperforming generic policies.

Complex scoring algorithms for KV cache compression are largely unnecessary: protecting the initial prompt and dropping reasoning tokens completely at random matches state-of-the-art accuracy with up to 43% higher serving throughput.

Test-time harnesses can nearly double the performance of weaker models, transforming how we think about capability transfer in AI.

Whisper's speech-centric training leaves audio-LLMs tone-deaf to music and environmental sounds, but a simple fine-tune can fix that.

Editing a rule in an LLM is not like editing a fact; you can't just tweak one layer – formulas live in early layers, instances in the middle.

Forget slow, end-to-end models: building real-time voice agents hinges on a cascaded streaming pipeline, as demonstrated by a new tutorial achieving sub-second latency.

Forget text prompts: vector prompt interfaces are the key to unlocking scalable and stable LLM customization.

Real-time voice agents can bypass slow vector DB lookups with a dual-agent architecture that pre-fetches relevant documents into a sub-millisecond semantic cache.

An 80B model that runs like a 3B? Qwen3-Coder-Next shows you can get competitive coding agent performance with a fraction of the active parameters, thanks to smart training.