Building general purpose robotics at Amazon FAR. @a16z Speedrun Scout. Previously founder @ pixelwise.ai / @AIatMeta / @Oxford_VGG PhD. nikitakaraevv.github.io
London, England
- Tweets 124
- Following 568
- Followers 3,245
- Likes 794
Nikita Karaev retweeted
Wrote done my thoughts on what's fundamentally new for vision when using agents --- obviously, apart from some hard problems that have now been nearly solved :)
1
9
54
8,718
Nikita Karaev retweeted
Introducing Unreal Agent:
An open-source harness with state-of-the-art cost efficiency
39% cheaper than Codex+Astra on Terminal-Bench 4.0 while maintaining performance
217
317
4,431
757,678
Vision goes agentic. From a bending spoon or a transparent glass to any articulated object, we can now reconstruct and track them in motion. Had a lot of fun working on this!
AgentSTAR is an agentic method for both reconstruction and tracking of articulated objects from monocular videos.
We can now track through rapid motion, severe occlusion and thin objects!
There’s no spoon but we can still track it 🥄
3
23
3,904
Nikita Karaev retweeted
AI creates a very strange economic effect. It can wipe out high-margin revenue and human jobs while making products and services dramatically cheaper, better, and more accessible. Billions in incumbent revenue may be replaced by millions in AI revenue (with most of it going towards compute) and a huge amount of consumer surplus that GDP barely captures.
As AI penetrates (and upends) more industries, this doesn’t necessarily mean people will feel poorer. Even with lower incomes, they may be able to afford far more—and receive much better healthcare, education, legal help, financial advice, entertainment, and eventually physical goods and services. Real living standards could rise while employment and measured GDP fall. At the same time we might see increased inflation, so Philips curve can also be broken.
But the order matters enormously. We need AI to democratize essentials first, and it’s not happening. A truck driver who loses his job won’t care that advanced mathematics is now nearly free if he still can’t afford eggs.
If AI lowers the cost of essentials fast enough, we could get more abundance despite less employment and less measured economic activity. If it disrupts incomes before it disrupts the cost of living, we get something much uglier: technological abundance alongside economic insecurity (which is exactly whats happening now).
Either way, classic macro stats need to be reworked. The Phillips curve weakens, GDP increasingly diverges from real welfare, and income becomes a worse proxy for what people can actually access.
I just fixed my dishwasher with the help of ChatGPT. A trivial task. I had been about to order a new one. So this software has increased the country's real wealth yet decreased the measured GDP. The main economic indicator is structurally incapable of registering the thing that actually makes people better off, namely the growth of knowledge
1
2
14
2,170
Nikita Karaev retweeted
I am seeing claims around LLMs, specifically Astra having made serious progress on robotics. The tasks that are demonstrated are simple pick and place tasks with parallel jaw grippers. LLMs can do planning, and the impressive demos in these tasks primarily show that. But robotics is also about dexterity and dynamics - which is why we need high frequency controllers/policies that can deal with torques and forces. So here is a simple challenge. Can you prompt an LLM to output the high frequency control commands for a legged robot in varying terrain e.g. ashish-kmr.github.io/ RSS 2021, CoRL 2022 (this is by now 5 year old technology, so I am not picking a particularly hard task). I am not questioning the usefulness of LLMs for high level planning or in agentically assisting a robotics researcher (we use them all the time!).
37
76
929
89,342
If you’re in SF, you can’t miss this!
Everyone is behind on AI.
Even if you’re in the field, every day, building, it’s just so easy to lose something the big picture.
Each year we gather a group of top thinkers and builders - researchers, founders, executives, engineers, and investors to do a 360 and try to catch each other up and fall down a few Rabbit Holes along the way.
Join us for the AI Rabbit Hole 2026: the AI Bubble Tea Party.
The lineup for this year is already insane: founders of top AI startups, researchers from top labs, execs from key infrastructure players, the most active and successful VCs.
Here’s a video from our last one:
3
7
647
Nikita Karaev retweeted
Christian Rupprecht explains their interpretability research in 3D computer vision, testing if (and where in the model) multi-view transformers like VGGT, DepthAnything 3, and DUSt3R use point/patch correspondences to make sense of 3D scene geometry.
5
100
678
74,780
If you’re a new grad and wanna start building - this is your sign. Apply.
a16z Speedrun Alpha, for pre-idea/pre-team/pre-everything founders
it's time to bet on yourself, and figure out your startup idea. 2026 is well underway, crazy stuff happening in AI, and you're building agents/apps/whatever every night+weekend. You want to start a startup but you're working or still going to school.
what if you're pre-idea, pre-product, pre-launch, and even a solo founder? You need time to cook
The Alpha Fellowship is for you.
alpha.a16zspeedrun.com/
details:
- $20K equity-free upfront to start building
- up to $250K investment when you finalize
- automatic final interview for a16z speedrun, with up to $1M investment
- 8-week, in-person experience with a kickoff retreat, founder AMAs, and small-group dinners alongside the a16z speedrun community
- targeted to early-career highly technical founders
- deadline to apply is March 6
We ALSO have a "startup track" for the Alpha Fellowship where you can get more founder experience by working for a portfolio company if you're not quite ready to found something.
The Alpha Fellowship places top early-career engineers into full-time roles at fast-growing a16z speedrun and Andreessen Horowitz portfolio companies. For future founders, we provide capital before a team or idea even exists.
We're looking for highly technical students and recent grads who don't want to wait to start building. Fellows take full-time roles at fast-growing portfolio companies - or, if you're ready to build now, receive capital to start your own company - kicking off with a two-month in-person fellowship. Fellows also have access to the a16z speedrun and EO Ventures communities and events.
...
If this is you, want to meet you. If you have people to introduce us to, that would be amazing too.
will have more to say, and lots of ideas coming up here. But excited to get this out! Excited to host y'all soon.
8
16
4,306
Let’s save Westminster and start building new castles in Britain again 🏰
Let’s go @ArkaSerezh!
2
367
Nikita Karaev retweeted
No book has had a big of an impact on my thinking. Every human on earth should read this, whether you’re technical or not. Thank you @naval for the recommendation, and thank you @DavidDeutschOxf for this masterpiece.
143
748
8,738
741,041
Interesting approach to 3D point tracking and dynamic reconstruction from @Oxford_VGG - looks very promising for robotics applications🤖
Congrats @SucarEdgar @EldarIsTyping @LaiZihang Andrea Vedaldi!
1
1
13
2,174
Nikita Karaev retweeted
What if we can simulate an *interactive 3D world*, from a single image, in the wild, in real time?
Introducing PointWorld-1B: a large pre-trained 3D world model that predicts env dynamics given RGB-D capture and robot actions.
🌐 point-world.github.io
from @Stanford @nvidia
31
218
1,280
278,613
Amazing work on 4D reconstruction and point tracking from DeepMind - congrats to @ChuhanZhang5 and the team! Super relevant for robotics applications.
A SINGLE encoder + decoder for all the 4D tasks!
We release 🎯 D4RT (Dynamic 4D Reconstruction and Tracking).
📍 A simple, unified interface for 3D tracking, depth, and pose
🌟 SOTA results on 4D reconstruction & tracking
🚀 Up to 100x faster pose estimation than prior works
10
1,181
🚨Big news - our team is joining the Frontier AI and Robotics (FAR) lab at @amazon to keep building the future of robotics. Excited to be joining such an inspiring team and to work with @peterxichen, @pabbeel, @rocky_duan, @akanazawa and many others at FAR.
19
23
489
83,182
Wow I didn't expect to receive so many DMs (400+), I'll go through them one by one.
The pattern I noticed so far, which wasn't obvious to me a year ago - most people jump straight into pitching their idea without saying a word about the team. The team is actually *the* most important part of the pitch.
21
146
18,969
Personal update - I'm now officially an a16z @speedrun scout. This means I can write you a 10k$ check within 24 hours.
If you're an early-stage founder or thinking about starting a company and have done something cool before - my DMs are open!
I'm mostly into AI/robotics, but always happy to support ambitious founders in any space!
171
70
1,863
226,197
Nikita Karaev retweeted
Thanks to AK for sharing our work!
🧩 Code: github.com/henry123-boy/SpaT…
🌐 Project Page: spatialtracker.github.io/
📄 The final version of our paper is coming in a few days — stay tuned!
3
18
101
30,228
Nikita Karaev retweeted
🚀 We release SpatialTrackerV2: the first feedforward model for dynamic 3D reconstruction and 3D point tracking — all at once!
Reconstruct dynamic scenes and predict pixel-wise 3D motion in seconds.
🔗 Webpage: spatialtracker.github.io/
🔍 Online Demo: huggingface.co/spaces/Yuxihe…
5
89
458
41,577