Building something new in Robotics and Physical AI! Adjunct Prof of CS at @Stanford, ex partner at @CalibrateVC & head of ML at @ToyotaResearch
Joined February 2012
- Tweets 3,179
- Following 1,493
- Followers 4,329
- Likes 9,898
237 Photos and videos
Feb 6
Physical AI is accelerating, isn't it? It's almost like something big is happening š The age of demos is coming to an end, and the age of real, useful, high-value work is upon us. Exciting times!!
3
2
32
2,290
20 May 2025
Hello from #ICRA2025 in sunny Atlanta š Looking forward to catching up with my robotics colleagues! I'll also be chairing a session on Thursday (ThET6) with Katherine Liu from @ToyotaResearch where we will present our work OmniShape tri-ml.github.io/omnishape/ See you there!
11
870
2 May 2025
3D is mainstream now: incredible progress in the past few years (e.g., on zero-shot performance). No reason to stay in 2D: elevate your vision š
š Details of the #CVPR2025 award candidate papers are out.
14 of 2967 accepted papers made the list, spanning 3D vision, embodied AI, VLMs/MLLMs, learning systems, and scene understanding. 3D vision leads with the most entries.
I collected the TL;DR, paper, and project linksš
11
1,035
Adrien Gaidon retweeted
25 Mar 2025
We do a lot of cutting edge research at the Stanford AI Lab, but really our main job is educating students. Here is a list of great SAIL Graduates of 2025, who are variously looking for academic and industry jobs! šŖ
ai.stanford.edu/blog/sail-grā¦
Compiled by @NikilSelvam Alex Nam @judyhshen
10
50
90
12,396
19 Mar 2025
The biggest bottleneck in robotics today? Data. Scaling up robot demonstrations is crucial, but we are still 5-6 orders of magnitude away from LLMs š± So how do we close that huge gap?
Ken's talk at #GTC25 is phenomenal and makes a strong case for scaling with Production Dataā¢ļø The evidence from @AmbiRobotics is clear, and we see that too at @CalibrateVC with startups like @BradPorter_ 's CoBot, @GrayMatterRobot, and more.
But to do that, you need a product, iteration in the field, continuous delivery of value, clear ROI... Robotics startups donāt just need funding - they need customers.
More broadly, Venture Capital is only a catalyst for the real reaction that happens in the field. The best AI companies know that and ship fast to get paid twice: in data AND in šµ. Thatās the unstoppable flywheel that happens when you build something people want.
That's not to say only production data matters. Robotics is so hard you need ALL the data: web data, sim, teleop, AND production data. But only one scales with customers. That's why we believe purpose-built robots are a massive unlock from today's foundation models AND the path to build tomorrow's even more general embodied AI.
PS: if youāre building in this space or thinking about jumping in, let's talk š
18 Mar 2025
2
12
2,669
18 Feb 2025
The demo-to-product gap is bigger than ever. AI is great for IA (Intelligence Amplification) but rough for AA (Artificial Agents). Intelligence and Autonomy have a complex relation --> closing the loop is key (mainly with humans). True for robots too!
9 Feb 2025
Me using LLMs for fun little personal projects: wow this thing is such a genius why do we even need humans anymore
Me trying to deploy LLMs in messy real-world environments: why is this thing so unbelievably stupid and dumb
2
459
Adrien Gaidon retweeted
18 Feb 2025
Calibrate Partner @adnothing is a judge at todayāsĀ 5th Annual GCEA Startup Competition. Best of luck to the #startups pitching their ideas! ceoceo.net/gcea-2025
1
1
260
27 Jan 2025
The magnitude of the Nvidia selloff is weird... Forget Jevons' paradox: does the market understand that these methods scale with compute? This was actually *reinforced* (pun intended) with R1! Or do they believe it is not going to get significantly better? š¤
4
425
27 Jan 2025
The AI battlefield is not benchmarks or abstract capabilities. We passed a blurry threshold where the battle is utility. Focusing on products is the right move. Agents is where it's at now, because autonomy is harder than intelligence.
2
339
27 Jan 2025
It is not entirely surprising, but still fascinating that suppressing language mixing in the training of DeepSeek-R1 hurts performance. Having learned Maths/CS in both French & English, I can definitely relate. Many multilingual friends also mixing languages at inference time!
2
6
559
25 Jan 2025
šÆ Soumith! 1-2 orders of magnitude cost gains is common once the "four minute mile" barrier falls. Remember ImageNet training in 1h then in 1min? Everyone is sprinting. This leaves a trail of low-hanging fruits that did not make sense to pluck on the 1st pass, but on the 2nd...
25 Jan 2025
i'm comically impressed that people are coping on deepseek by spewing bizarre conspiracy theories -- despite deepseek open-sourcing and writing some of the most detail oriented papers ever.
read. replicate. compete.
don't be salty, just makes you look incompetent.
1
2
751
12 Jan 2025
At the Global AI Pitch Summit in sunny Sunnyvale today and tomorrow! Excited to hear from the many startups and keynote speakers. Hit me up if you want to chat!
1
2
428
23 Dec 2024
OpenAI's o3 results got everybody very excited (and possibly worried). Following the "AGI Asymptote" post by @BradPorter_ , I wrote down my own thoughts about what it means, and how to avoid the trap of ELIZA Bonini and the AI turtles: adriengaidon.com/posts/2024/ā¦
TL;DR: ML reasoning systems are going through an exciting phase transition that escapes our understanding; we neither can nor want to stop it, but this requires an AI Safety approach similar to autonomous driving (engineering human accountability) to avoid the pitfalls of AI anthropomorphization.
What do you think?
1
1
5
527
23 Dec 2024
Brad's post that got me thinking: linkedin.com/posts/brad-portā¦
Some other sources of inspiration beyond Brad: @random_walker and his AI snake oil blog (and book!), @natolambert and his interconnects blog, @rao2z , and Gill Pratt. Curious what y'all think about o3 & AI safety!
197
19 Dec 2024
Another banger AI Snake Oil essay š Go read it! I have two high-level take-aways on forecasting and ML systems:
1) Predicting the future is hard, even for insiders! Famously, it's better to just build it. But in an attempt to get an edge / move where the puck is going (there are no moats), some tend to extrapolate too far based on higher order moments (e.g., acceleration vs position). This amplifies the noise in high uncertainty situations. That's why common-sense slaps across the face like this essay can bring everyone back to their senses and go after the real potential, not mirages. Increasing accountability would also be nice ("conduct a systematic analysis of all predictions about AI made in the last 10 years by prominent industry insiders"), but very few people have the courage to do that (@rodneyabrooks comes to mind with his legendary self-driving scorecards rodneybrooks.com/predictionsā¦ Looking forward to the January 2025 one!).
2) Repeat after Arvind: "models themselves are just a technology, not a product". Models are often a small (yet crucial!) part of products -- yes, even big models or agentic systems! -- per the timeless "Hidden Technical Debt in Machine Learning Systems" proceedings.neurips.cc/paperā¦
19 Dec 2024
New AI Snake Oil essay: Last month the AI industry's narrative on scaling suddenly flipped. This has left people outside AI confused. What changed? Is AI capability progress slowing? We look at the evidence.
By me, @benediktstroebl and @sayashk.
aisnakeoil.com/p/is-ai-progrā¦
-----
After the release of GPT-4 in March 2023, the dominant narrative in the tech world was that continued scaling of models would lead to artificial general intelligence and then superintelligence. Those extreme predictions gradually receded, but up until a month ago, the prevailing belief in the AI industry was that model scaling would continue for the foreseeable future.
Then came three back-to-back news reports from The Information, Reuters, and Bloomberg revealing that three leading AI developers ā OpenAI, Anthropic, and Google Gemini ā had all run into problems with their next-gen models. Many industry insiders, including Ilya Sutskever, probably the most notable proponent of scaling, are now singing a very different tune:
āThe 2010s were the age of scaling, now we're back in the age of wonder and discovery once again. Everyone is looking for the next thing,ā Sutskever said. āScaling the right thing matters more now than ever.ā (Reuters)
The new dominant narrative seems to be that model scaling is dead, and āinference scalingā, also known as ātest-time compute scalingā is the way forward for improving AI capabilities. The idea is to spend more and more computation when using models to perform a task, such as by having them āthinkā before responding.
We make four main points in the essay:
aisnakeoil.com/p/is-ai-progrā¦
ALT Declaring the death of model scaling is premature. Regardless of whether model scaling will continue, industry leadersā flip flopping on this issue shows the folly of trusting their forecasts. They are not significantly better informed than the rest of us, and their narratives are heavily influenced by their vested interests. Inference scaling is real, and there is a lot of low-hanging fruit, which could lead to rapid capability increases in the short term. But in general, capability improvements from inference scaling will likely be both unpredictable and unevenly distributed among domains. The connection between capability improvements and AIās social or economic impacts is extremely weak. The bottlenecks for impact are the pace of product development and the rate of adoption, not AI capabilities.
1
7
24
5,978
18 Dec 2024
2024 was a massive year for Calibrate Ventures, our portfolio companies, and the deep tech ecosystem as a whole! And this is just the beginning of putting data to action in the real world. Excited for whatās ahead in 2025 and canāt wait to partner with more visionary builders! š
18 Dec 2024
From groundbreaking ideas to meaningful connections, 2024 challenged and inspired us in equal measure. Through it all, weāve been driven by the resilience and ingenuity of #founders and their teams turning moonshot ideas into real-world impact.
Explore the bold ideas taking flight across our #portfolio and what we're excited for in 2025 in ourĀ #YearInReview from @JasonSchoettler andĀ @nbvc4: calibratevc.com/blog/2024-atā¦
1
3
572
9 Dec 2024
On my way to #NeurIPS2024! Please reach out if you want to chat, and check out our paper on streaming multimodal video understanding with my awesome @StanfordSVL colleagues @CristbalEyzagu2 , Eric Tang, Shyamal Buch, @jiajunwu_cs , and @jcniebles š
6 Dec 2024
š¤ āHey assistant, don't let me forget my card at the ATM!ā
š¶ Are you as forgetful as I am? Imagine you could ask your smart glasses to give you a reminder the next time a certain event happensā¦ wouldnāt that be amazing?
1
10
1,069
Adrien Gaidon retweeted
21 Nov 2024
We revealed our new cobot, Proxie, yesterday. Today we want to share more about the work Proxie is doing in the world.
youtube.com/watch?v=2Pr9YP0vā¦
1
3
35
2,576
20 Nov 2024
CoBot's Proxie revealšš Not just the design (human-level capabilities without being a humanoid!), but how useful it is in the field today (5 million pounds moved already!!).
In an era fraught with overpromise and demoware, you don't see this kind of news often: new robot, advanced embodied AI, and, most importantly, real commercial traction. Everyone has that dream, a few have tremendous potential, but almost no one can back it up with tangible results today.
Crossing the demo-to-product gap (a.k.a. the valley of death) is a feat very few accomplish, especially in robotics, and especially with robots that need to amplify people vs replace them. People have a very high bar for their metallic colleagues.
Congrats Brad and the wonderful humans and robots at CoBot on this massive milestone. So happy to be part of your investors with @CalibrateVC and excited for the journey ahead: this is just the beginning!!
20 Nov 2024
Excited to introduce Proxie, our first cobot, to the world today!
3
604