Will Knight

robot, initiate "chose a chocolate from the box mode..."

@Hiraweb3 @PrimeIntellect @adaption_ai really well! highly recommend prime intellect!

@willknight @PrimeIntellect @adaption_ai love seeing more control in the hands of users how'd the fine-tuning go?

For my newsletter this week, I tried fine-tuning my own models using tools from @PrimeIntellect and @adaption_ai. Most of the big labs are using AI to build powerful models, but the same approach could also give people greater control over their own AI. wired.com/story/frontier…

@jietang multimedia capabilities would be nice. also the ability to create your own research harness!

Any new features we must have in the next version of glm?

@alexocheema I dmed you about trying it out!

Anyone at AIE badge pickup (Moscone) want an exclusive sneak peak of local.ai? I have it on my laptop here.

@DJiafei yes, good point. i think there are some very exciting things happening in manufacturing these days, I need to learn more!

@willknight Totally agree! Needs to happen in parallel for us to see the first example or signs of real world deployment!

Robotics still seems to be very much an academia game, just today alone there are 174 papers on ArXiv, and most of them are on VLAs and WAMs.

I spent the past week and a meeting with a lot of Chinese AI experts. American AI bros have more in common with them than they think. wired.com/story/ai-arms-…

The US needs to figure out how to accelerate its own robotics industry rather than just impede foreign companies. American manufacturing badly needs a reboot.

@lzwjava @zacharylipton @GoogleDeepMind @JeffDean @demishassabis i think i moderated a few sessions, including one featuring Zach.

@willknight @zacharylipton @GoogleDeepMind @JeffDean @demishassabis speaker or participant? ha

After countless yrs at @GoogleDeepMind, I've decided to join [company]. Impossible decision. Grateful to @JeffDean & @DemisHassabis for the mentorship & late-night scientific ruminations. GDM taught me how to eat, breathe & think. Special place. Excited for their next moves 🚀

@lzwjava @zacharylipton @GoogleDeepMind @JeffDean @demishassabis i was there as well.

@zacharylipton @GoogleDeepMind @JeffDean @demishassabis Zachary, you went to china Beijing to give an AI talk around 2017, We met shortly, and connected with wechat . Now I learn AI part-time since 2023. Time flies, thank you for leading the world! And best wishes to your next journey. The jiqizhixin conference.

@j2bryson i don't know to be honest. but there are still more AI and robotics people here I think...

@willknight ha, sort of! 17K followers there, do you really not get more click throughs there than here? Or do think the musk machine lies about how many impressions our posts are getting?

I just learned a lovely Chinese phrase for dystopian: 反乌托邦的 (Fǎn wūtuōbāng de) meaning literally "anti-utopia".

@clankrmedia This is good but i don’t think it’s proof the ChatGPT moment is here.

The GPT-4o moment for humanoids might finally be here. And yeah, sorry in advance for the rickroll. OMG runs a Unitree G1 off one brain that natively takes language, audio, and human motion. That "Never Gonna Give You Up" dance isn't the flex; one model fluent in every modality is. Here's the shift. Most humanoid policies are one-trick. Train per skill, hand-tune the rewards, repeat. The rest just replay a fixed motion you feed them. OMG instead works like a biological motor system. A "brain" that turns intent into future motion. A "cerebellum" that reactively runs it on the robot. The brain is one diffusion model. Language, audio, a reference pose, or any blend goes in. A robot-ready G1 trajectory comes out, live. New inputs attach through zero-init adapters. They start at zero, so the pretrained motion prior carries over intact instead of getting scrambled. That's how they bolted on VR teleoperation as a brand-new modality, reusing the same brain. And it behaves like a foundation model. Bigger backbone, cleaner motion. Finetune on 1% of new data, nearly match a model trained from scratch on 100%. Compose language + audio at inference for combos never seen in training. 1000+ hours of motion, all retargeted into one G1 body. One brain that scales. We keep racing to build stronger low-level controllers. OMG's bet is that the real bottleneck is the brain mapping human intent to motion. Congrats to @KnightNemo_, @li_yitang, @ShaotingZ38103 and whole team!

@alexocheema This is the way

@j2bryson Stanislaw Lem makes sense now…

@willknight I learnt from a Polish postdoc that the word for dystopia & utopia is the same :-) BTW why talk to 200 people here instead of at least that on bluesky? Or are you there and I haven't found you? (I'm just checking my "reality" list to find whether The Strait of Hormuz is open)

Dario is in a very Chinese time of his life rn

@kevinsxu What a time (in AI) to be alive…

😂😭🇺🇸🇨🇳