“The talent level required to train a massive model with high FLOPS utilization on a GPU grows increasingly higher because of all the tricks needed to extract maximum performance.”

How Nvidia’s CUDA Monopoly In ML Is Breaking – OpenAI Triton & PyTorch 2.0 https://t.co/zrOtrOBKVX

With the NeRF technique, you take just a few pictures of a 3D space, and the #AI generates all the views you did NOT capture with the camera (aka view synthesis).

Then you put those views all together as frames of a video and you get a very cool AI-powered commercial. https://t.co/RuRvfrlpdh

Think about what this could do to, for example, the #fashion industry if you take shots of runway models on a catwalk.

@Merzmensch If you are desperate to try it, you might want to change your location via a VPN and restart the app. Maybe that will do.

@Merzmensch Not in the UK. I just took the screenshot on my mobile as you can see from the 5G indicator and the ratio.

Did you update the app recently? Did you kill/reopen it (probably won’t make a difference)?

@Merzmensch https://t.co/3iOO7NeHmK

Interesting thought. https://t.co/va4qvhQcnm

@Merzmensch It’s not just you. It’s everybody. Many posts of the people we follow directly appear in the “For You” but not in “Following” tab.

It turns out that people love the “Move fast and break things” culture only when it doesn’t impact the things they use. Like Twitter.

RT @ritholtz: Is Crypto like a Security or a Bank?

Think carefully before you answer, says @matt_levine

https://t.co/OCQ7xYknbR https://…