“You’ll literally have streaming data … going into like hundreds of models … and have it all be served to the user in sub-20, sub-50, sub-100 milliseconds”
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) · Your Ranking Model Is Optimizing the Wrong Thing with Zoe Weil
Play from 50:08
50:08
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
Route streaming browser events to hundreds of specialized models that each reconstruct and output data, then feed into a final ranker
sub-20 ms, sub-50 ms, sub-100 ms latency
Said nearby on the same subject
“We have usually hundreds of terabytes of hard drives on a job site now… using computer vision to turn it into an ongoing data stream, not just of pixels, but of events, activities”
CB
Chris Bayes
Digital Builder ·
Why it workedhundreds of terabytes of video per job siteShare this moment
“We trained the final model on the SpaceX cluster ... a few weeks of pre-training, then synthetic data NRL and then shipping it”
ML
Misha Laskin Beam
No Priors: Artificial Intelligence | Technology | Startups ·
“With this video recognition, VLM models, you're able to just very quickly take a live streaming video, convert it into a text narrative, and then match that against your community guidelines”