Andrew Carr 🤸
@andrew_n_carr
Followers
24K
Following
38K
Media
1K
Statuses
10K
co-founder leading science @getcartwheel co-founder advisor @arcade_ai Past: Codex @OpenAI, Brain @GoogleAI, world ranked Tetris player
RuntimeError: shape is invalid
Joined July 2015
As a math loving computer scientist, when people suggest my work isn't theoretical enough - I just repeat "Hilbert space" over and again until they leave.
13
25
744
It's a bit unbelievable how composer 1 is still a daily driver for me.
1
0
3
i'll say what we're all thinking - GDPval is a great name for an eval
0
0
10
I saw the android <> airdrop announcement a while back. Is that real? How do you get it?
1
0
4
I'm excited for the day when chat gpt version number and chad gdp are the same
1
0
2
Reliable backup when you need it most. Keep your home running during severe weather.
0
75
377
You should train a model called "game changer" so it's guaranteed to be a game changer
0
0
7
I'm pretty sure the "AI Mode" / "Generative Search" model is something unreleased. Time to figure out it's mmlu-pro score
11
1
108
Devstral 2 for agentic
seems like the open stack is glm-4.6V-Flash for multimodal olmo 3 32B thinking for math rnj-1-instruct for agentic
0
2
48
Huge kudos for the transparency here. Beating Kimi k2 as well is stellar
Introducing the Devstral 2 coding model family. Two sizes, both open source. Also, meet Mistral Vibe, a native CLI, enabling end-to-end automation. 🧵
3
17
302
Have we gone too far?
Are images and text enough to control 3D generative models? SpaceControl shows there’s more: we turn 3D drafts into detailed, perfectly aligned 3D assets ✨🚀 📜: https://t.co/RzN8nPfeNI 🌍: https://t.co/4xF83r9Ehu
2
1
10
Still my fav model and cli. Gemini 3 pro + cli also is the first agent/terminal combo that doesn't routinely make my blood boil while using it.
After a couple days of heavy use, tldr: it’s a good model and gemini cli is very good. The pair might be the best for terminal-based agents. But it’s not perfect. Pros: I never hit any limit for gemini 3 pro and used it for multiple 3-5+ hour sessions. I will get my money's
3
2
31
seems like the open stack is glm-4.6V-Flash for multimodal olmo 3 32B thinking for math rnj-1-instruct for agentic
7
11
206
My wife's family owns a grocery store. They've started to digitize. Every post on Facebook has two likes. Me and my brother-in-law. 💪
2
1
44
First they came for the Software Engineers And I did not speak out Because I was not an Engineer Then they came for the Photographers And I did not speak out Because I was not a Photographer Then they came for the Hydroneer players And there was no one left To speak out for me
1
0
7
Odd how the things that make parenting the hardest will be your child's greatest strengths through life.
0
0
4
One of the biggest challenges in video generation is temporal character consistency for videos longer than 8s. The Live Avatar work from Alibaba proposes a perfect solution for this. By dynamically updating the sink frame’s RoPE to preserve relative positioning they are able
3
25
283
Warm up, stable, decay stays unbeatable. 8T tokens, variable batch size, big vocab, impressive benchmarks - what's not to like?!
Today, we’re excited to introduce Rnj-1, @essential_ai's first open model; a world-class 8B base + instruct pair, built with scientific rigor, intentional design, and a belief that the advancement and equitable distribution of AI depend on building in the open. We bring
1
6
120
It's amazing how many language models refuse to believe that netflix bought warner bros for $82B
10
1
201
This is the falsest statement ever written by man
Every man learns for himself the uselessness of owning an ipad
0
0
7