Yiren Lu

Tried out @googleaistudio for the first time yesterday (for a video captioning task) and it's actually great! - fast onboarding, they give you the developer key right there and you only have to install one python library, no random google cloud secrets files - super long context

This actually makes a lot of sense hmm

Josh Whiton @joshwhiton

Friend just asked me to build a locally running AI rig for him in case of a disaster. In an emergency he wants to have food & water stores, generators, and AI. Makes sense. A local AI model could talk you through water de-sal, electrical work, hunting and foraging, first aid...

RT @daniel_d_kang: I am recruiting PhD students this cycle! If you're interested in: - AI agents and security - both AI agents for cybersec…

I fed @notebooklm_pods the Wikipedia page of Empress Cixi and asked it to generate a podcast in the style of a documentary, focusing on her rise through the imperial harem. Here's the result: https://t.co/8oMZBN2dEu The good: - the speech synthesis is pretty extraordinary. It

I really really want to like Claude’s computer use feature but I asked it to put together 30 wedding registry items from West Elm and Crate and Barrel and it hallucinated a bunch of links. Then I asked it to find 10 wedding florists in the Charleston area and ask them for quotes

What I liked most about SAM2 from Facebook for video segmentation: - worked out-of-the-box on the first sample video I gave it! - lets you click on the object you want to segment out on the first frame and then it automatically propagates out to the rest of the video frames -

Charles 🎉 Frye @charles_irl

A bit late to the party, but: @yirenlu92 and I just added a code sample for running SAM 2 on @modal_labs! https://t.co/fs5QkA75kE

Have been using @cursor_ai the last few weeks and really like it, although sometimes when you ask it to make edits to code it seems like it just runs out of output tokens and omits a bunch of stuff. Definitely have to be careful blanket accepting its suggestions. I primarily do

ok does this support Chinese because if so I need to build an English-Chinese translation device for my exclusively Chinese-speaking nanny to communicate with my husband.

Erik Dunteman @erikdunteman

The Moshi speech-to-speech model by @kyutai_labs, hosted on @modal_labs, absolutely cranks! Near instantaneous voice chat AI, thanks to bidirectional streaming. Whole startups could be built on this stack. https://t.co/wORwirpQUx

Really impressive demo of the Ultravox text-to-speech model here: https://t.co/qAEGJB0XcL A lot of the open-source text-to-speech libraries aren't really fast enough for real-time AI conversations, but this one is pretty damn close.