Ted Xiao
@xiao_ted
Robotics and Gemini @GoogleDeepMind. Posts about frontier models, robot learning, and scaling. Opinions my own.
📢 First contact between a frontier model and robots! Gemini Robotics is a SOTA generalist Vision-Language-Action model bringing frontier model intelligence to the physical world. It's an extremely capable model enabling dexterous, steerable, and general robot control. 🧵⬇️
““In our main experiments, a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a "student" model trained on this dataset learns T” What?? 🤯 subliminal-learning.com
Bonus: Can *you* recognize the hidden signals in numbers or code that LLMs utilize? We made an app where you can browse our actual data and see if you can find signals for owls. You can also view the numbers and CoT that encode misalignment. subliminal-learning.com/quiz/
Gemini Deep Think achieves gold-medal level results at the IMO 2025🥇based on official grading results. The unreasonable effectiveness of extended autoregressive “thinking” continues to push forward. Congrats to the whole team!
Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! deepmind.google/discover/blog/…
Unhacked benchmarks with no leakage are increasingly rare. Math and coding competitions are a great way to fairly evaluate reasoning models! …all the top frontier models are already better than I ever will be on the hardest competition math problems 😅
LLMs for IMO 2025: gemini-2.5-pro (31.55%), o3 high (16.67%), Grok 4 (11.90%). matharena.ai
The most significant impact of superhuman AI may not be economic or technological, but spiritual. How do leaders of faith view AI today, and how will this relationship evolve as it reaches critical thresholds of intelligence and ubiquity? On one hand, it's easy to frame strong…
This is the most polished short film powered by AI-generated music, images, and video that I have seen to date. Pretty astounding what systems like Veo and Sudo can do in the hands of talented creators 🤯 youtu.be/gx8rMzlG29Q?si…
Our Aloha robot joined us in-person at RSS for a live demo of our new Gemini Robotics On-Device model, come check it out!
Come by the @GoogleDeepMind booth at @RoboticsSciSys conference in LA! We’re demoing Gemini Robotics On-Device live, come check it out