Despite being a newbie to nuclear, it's obvious it will be a key form of clean, abundant energy to power the new economy. I just released the full conversation (link below) with Liz Muller from @DeepFission . They're building micro-nuclear reactors 1-mile underground.

interesting take on nuclear waste. if you use nothing but nuclear power your entire life, it would fit in about one soda can.

if you understand someone's incentive structures, you will better understand their behaviors and find more common ground.

Link to full episode:

@garrytan @Noahpinion 1) American factories skew smaller than Chinese & others so they have less justifable workloads 2) American robotics delivery has generally relied on very expensive system integrator networks that are incentivized towards customization, cost-plus project work, and vendor lock-in

I know very little about nuclear energy so I decided to go nuclear this week. I sat down with Liz Muller from @DeepFission who's building modular, micro nuclear reactors one mile underground. A little tasty-poo below. full episode drops later this week on Going Direct.

Full conversation is up at Going Direct.

3/ Model size and # of parameters is important in robotics, but balancing inference speed w/ model size is critical. Much more than with LLMs.

2/ Context length is critical if we want to robots to have the ability to reason and do complex tasks. Much easier said then done…

1/ Interference speed is much more critical in robotics than LLMs or a chatbot.

Any robotocist will tell you many gaps still exist when it comes to training robots. I discussed three with Karol and Kevin from the Physical Intelligence team: 1. Inference speed 2. Context length 3. Model size (limited amount of data) 🧵 below:

4/ Karol and his team want it to make it really easy to build in robotics where intelligence is waiting there for you.

3/ Large tasks have been solved with specialized/ expensive robotics. Now we have to tackle a long-tail of tasks that do not work with traditional techniques.

2/ Way too much friction right now if you want to start a company to solve a problem in automation or robotics.

1/ Robotics ecosystem barely exists today

Entire ecosystems will be required to unlock physical intelligence/generalized robotics. Current state of industry below from my conversation w/ Karol + Kevin at @physical_int 🧵

I joined 1X because Bernt & Eric asked me if I could help them make NEO into something simple enough for anyone on earth to use and give it to the world. It felt like having a couple strangers come up to me and offer me a chance to change the world (well it was). I've never

World models with multi-modal data are critical to making low-cost, generalized robotics a reality. I discuss this w/ @hausman_k and @kvablack from @physical_int on Going Direct. Full episode below. YT: