Andrew Carr 🤸
@andrew_n_carr
co-founder leading science @getcartwheel co-founder advisor @arcade_ai Past: Codex @OpenAI, Brain @GoogleAI, world ranked Tetris player
5.2K Following 28.7K Followers
we hire some of the worlds greatest artists, they are now all (100%) using codex in their workflow somehow. just in the last two weeks, unprecedented adoption
They probably don't quite have a billion users yet
Imagine how much money typesafeAI made with the jev launch. Like it must be double digits at this point already
Ironic that muse is blowing up here and not grok bot
My whole feed right now...
Unrelated weekend post. I went camping with my dad and my son last night. It was awesome. We got soaked, threw rocks, built a big fire, ate meat, pretended to sleep in the tent.
It was great. Truly a moment in time with healthy parents and young kids. But it's a good one.
Show more
1. Get nemotron and olmo datasets
2. Format them with dsv4.1 flash into jev shape
3. Take needle 3 arch, scale up
4. Get O(1k) gpus
5. Train your own jev
2 years later, you've got your own
it used to be that you talked about a product with friends, and it later showed up on your instagram timeline
now you'll talk about your api key and muse will get it
> Opus: "It won't work till we have the API key, but do NOT share it with me!"
> Codex: "We'll need the API key to test, but don't share it here."
> Muse Spark on contributor mode: "One thing left is the API Key. Paste it here and we're good to go. Just paste it bro. Paste it"
Show more
there's a whole universe of *ev names out there just waiting for people to take
their jev is smart, your open jev is not
this is such a breath of fresh air in my feed full of seedance anime
Animations layers 👌
Actions + Nonlinear Animation in Blender.
I always try to use different actions to organize my animations, and in Blender is super cool we have this NLA.
#animation# #chainsawman# #blender# #blender3d#
Show more
this figure from the deepseek v4.1 flash paper is a must study.
look at the extremely sharp increase in quality after extending context length to 1M tokens.
you see a similar gain in the mimo-v2.6 rl graphs as well.
agents are context hungry
Show more
In this house, we pay for tailscale
Can we all just pause for a sec and reflect on how fucking important Tailscale is, and how nobody ever talks about them? I don't think I've ever seen a more useful service with a more invisible corporate footprint.
Show more
It seems like we're getting another Qwen Image!!
the hot new word should be "kinematic monocular object capture from any camera" but pacing works too
pacing is the hot new word in San Francisco. no one knows what it means but it gets the people going
Ah!! We've been trying to get this working for months.
Basically, we can now capture a person and the objects they're interacting with in 3D from regular video. Motion for both, plus approximate object shape.
I've been excited about this for a while because so much of what people do involves stuff. If you want a robot to learn how to use a tool, or to understand someone's tennis swing, you probably want to know what the tool or racket is doing too.
Same goes for props in animation and everyday tasks in occupational therapy. There's a lot here.
Still imperfect, but this week it started working well enough that I wanted to show you.
Show more
math PhD is perilous due to AI?
my friend it was perilous before too
the weird part of this MiMo graph is that the judge and probe disagree 60% of the time.
and for pro, disagreement goes up during training.
would love to know how much of that is harder-to-judge behavior vs the model getting better at fooling the judge.
Show more
prompt: 3 people sitting around and laughing.
Text to motion is getting a massive upgrade
1. multiple characters
2. faces
3. hands
4. much better prompt adherance
5. higher fidelity and subtle acting performance
all in 3d, coming soon!!
Show more
prompt: "Three people sitting around and laughing" 👀
Blackwells have arrived
