Asked why humanoid robots still can't load a dishwasher after $6 billion in funding, Fei-Fei Li says the number is too small: Emily Chang: "Funding for humanoids hit $6 billion, but they still can't load my dishwasher as fast as I can. They still can't go get my Amazon... packages. Will world models, world labs, close the gap between hype and reality?" "That's a loaded question, Emily." "First of all, robotics is gonna be one of the most important revolution in human industrialization." "$6 billion is too small, right?" "If you look at self-driving cars investment, if you look at language models investment, it took way more than six billion dollars." The $6 billion is Chang's total for humanoid funding, not Li's own measurement. Chang also says World Labs, which Li co-founded, has raised $1 billion of its own. - Fei-Fei Li (Fei-Fei Li), co-founder and CEO of World Labs and Stanford professor, with Emily Chang on The Circuit (Bloomberg TV), August 19, 2026.show more

"The money is too small" is the answer every unsolved problem gives when asked why it's unsolved. Self-driving is an odd comparison to reach for, since it absorbed far more than $6B and still mostly isn't here. LLMs scaled because the internet had already written the training data for free. Nobody pre-recorded a billion dishwashers being loaded. Capital buys compute. It doesn't buy a decade of physical data that was never collected.

How much are you willing to spend for a humanoid robot to do simple tasks? It does not make any economical sense. It is true that robotics will lead to a level of industrialization we have never seen before, but it will be specialized robots not generalizable humanoids.

一旦机器人可以送快递，做家务或者盖房子，每一个领域都是价值几万亿的市场，所以现在60亿的投入完全不值一提

Worth naming the actual bottleneck. There's a three-stage way to think about this: Era 1 was code-driven automation, explicit programmed steps. Era 2 is today's statistical/world-model AI, pattern-matching from massive data, which is what self-driving and LLMs scaled with. Era 3 is intent-native: a system that infers what you actually want and executes it in a messy, unstructured environment without needing that exact scenario in its training data. Loading a dishwasher isn't hard because of insufficient data volume, it's hard because it requires acting on ambiguous, real-time intent in physical space. More Era 2 funding buys better perception and simulation, not that capability. The $6B question isn't 'is it enough,' it's 'enough for which paradigm.'

Remember the Six Million Dollar Man? Now we have the Six Billion Dollar Man. It’s built for efficiency and not speed.

It took billions of years of evolution to create a species that learns you shouldn’t drink from the same water you put waste in. We’re a few years into actual spending on AI and it can manage drone swarms and hypersonic missiles. Give it a second. The dishes will get washed just fine eventually.

Analogize Analogize Analogize

It’s not money. You can put a trillion dollars to establish a human colony on mars and still fail.. you could have put Bullions of dollars solving the navir stokes - and still fail…people dont understand that AGI or general purpose robotics doesn’t exist.. its a math problem.. not money problem

باقي خاصهم يخدموا بزاف باش يوصلوا لهاد المستوى، الصبر وصافي!

🤖🚿 Can we get funding to teach them how to multitask (and a decent human sense of humor)?

Humanoid robots connected to a central AI are what people actually fear.

Si on a un robot il me semble inutile d’avoir un lave vaisselle...

What if the computational primitive itself should be continuity of evolving organization rather than successive representations of state? It explains why simply adding a richer “world model” doesn’t necessarily cross the boundary you’re interested in. A world model can become fantastically detailed while still being a model of states/things and their transformations. CAPS (Continuity Attention Protocol System) instead arose from the operational problem of maintaining what remains consequential through transformation. MoM (Morphology of Morphology) later sharpened the complementary observational question: what information resides in the morphology of that transformation itself? A paradigm is needed away from the current metric-centric one.

Interesting

Six billion dollars and they still can’t load a dishwasher. Fair enough. But I’m old enough to remember when computers couldn’t recognize a cat. I wouldn’t bet too heavily on the dishwasher remaining humanity’s competitive advantage. We’re building the brains now. The bodies are coming. And once the two meet, AI will have a physical way to reach out and touch our world, including the dishes.

本质上自动驾驶和人形机器人在物理世界的状态都是一样的，都是探索人类世界，但自动驾驶简单很多，它自己负责开车，只在道路层面活动，而人形机器人的行动范围和要承担的工作，复杂程度则是指数级的。人类在制造人形机器人的时候，可以借助自动驾驶的经验来更好完善人形机器人的制造