5 stories that day · 1 model · 1 open source · 1 research · 1 trending · 1 GitHub trending
Published daily at 7 PM ET · Updated · RSS
ModelTop story
Tavus says Griffin, a new kind of model, holds live face-to-face video calls, reading expressions and pauses as well as words. Only selected testers can try a smaller version, Griffin-Lite, for now.
Open source
A test bed that grades AI agents on what they actually change in a database, not on what they say. Microsoft and Hugging Face ran 507 business tasks 20 times each to see if agents succeed every time.
Microsoft · Original source (opens in a new tab)
Research
RIDE measures how reinforcement learning shifted a teacher model's internal signals, then trains a student to move even further in that same direction.
465 upvotes
Trending
Strata runs a 125-billion-parameter model on a gaming PC (opens the original source in a new tab)A free, open-source program that runs Qwen3.8-Flash-Next on Windows or Linux PCs with an NVIDIA or AMD graphics card of 12 GB or more. Its makers measured 94 tokens (word pieces) a second on an RTX 5070.
542 points on Hacker News
Strata · Original source (opens in a new tab)
GitHub
ponytail (opens the original source in a new tab)Instructions that push an AI coding agent to write as little code as possible, on the idea that the best code is code you never wrote.
DietrichGebert / ponytail1,894 stars today154,809 total starsJavaScriptas of 4 Oct
Notable stories from the past few days. Each shows its own date.
