OpenTube

AI Didn’t Run Out of Data - It Ran Out of Reality

House of El: AI

145,166 views

Physical AI is not failing because models are choking on “junk data.”
It is hitting a deeper constraint: the real world has not been instrumented at the scale needed to teach machines how reality actually behaves.

🤖 Text-based AI could train on the internet, but robots, self-driving cars, and world models need sensor data from physical reality.
📡 Physical AI requires cameras, motion sensors, pressure sensors, accelerometers, thermal data, and real-world interaction records.
🚗 Tesla’s autonomous driving advantage comes from billions of miles of fleet data — not just better simulation.
🏭 Manufacturing, healthcare, smart cities, and energy systems are quietly building the sensor infrastructure AI will need next.
💸 IoT deployment is growing rapidly, but it lacks the hype and funding attention that large language models receive.
⚠️ Simulation alone cannot capture the messy edge cases of real physics, materials, friction, weather, and human environments.
🧠 The real bottleneck is incentive alignment: people will accept sensors when they solve immediate problems, not when they merely feed future AI models.

🧠 Help Shape the New Course: AI Fluency for Thinkers
I'm building a course to help people navigate AI with clarity and confidence.
Take 2 minutes to shape the curriculum and get early access:
👉 https://houseofel.net/ai-course/

☕ Support the channel on Ko-fi:
➡️ https://ko-fi.com/houseofel

#AI #ArtificialIntelligence #PhysicalAI #Robotics #WorldModels #Sora #OpenAI #Tesla #AutonomousDriving #IoT #Sensors #DataInfrastructure #MachineLearning #ComputerVision #SmartCities #AIInfrastructure #TechAnalysis #AIFluency #HouseOfEl

20 Comments

@transformational_experience 4 months ago

To be precise: AI ran out of descriptions and interpretations of reality created by human intelligence to integrate into statistical models.
Now AI begins to learn how to better hallucinate from another AIs fever dream.
While we're wasting irreplaceable resources on a bunch of psychopaths bid to become immorally rich.

@ApocryphalDude 4 months ago

And some of us are poisoning the well intentionally.

@looseycanon 4 months ago

No, we're not cooking. Monitoring can and even now IS being abused beyond measure. This infrastructure must never be allowed to be built. A human has the right to privacy and this completely destroys it.

@ntw3002 4 months ago (edited)

~~'We can cut your energy usage by 20% with AI'~~ 'We can double your energy usage but divert the cost to households in districts with data centres'

@DrawingSerotonin 4 months ago

Am I the only person who feels like AI is a waste of billions of dollars and years of effort to literally “reinvent the wheel”. All of the damage to the environment for data centres, all of the damage to the economy, the loss of jobs, all the deaths that self driving cars caused, all to build something that already exists … people. And for what? It seems to me that this is all in aid of a return to slavery. Businesses want robotic employees who can think and plan and behave just like people, but who they don’t have to pay or treat ethically. How is this whole industry not just mindlessly evil on every level?

@RealMomar27 4 months ago

Or... you can hire a human, we come with all kind of sensors incorporated.

@markd5067 4 months ago

Garbage in garbage out

@leamael00 4 months ago

I'm part of the team producing data for factuality training on Gemini.
It's pure theater. It's very obvious that no one in the production chain cares are data quality. There's an open secret, which says: so long as Gemini sucks, we have a job.

@user-wh2nm2sv9w 4 months ago

No, for the sake of privacy, we don't need that many sensors.

Privacy is much more important than ai.

@alanfarlie759 4 months ago

One of the fundamental problems with the idealistic ideas about data gathering is that the people in charge are billionaire sociopaths with disturbing political and societal opinions.

@davidhache3642 4 months ago

Depressing to hear ppl have to track all their lives for a few bucks to train robots. Dystopia.

@snarklie962 4 months ago

It makes me bonkers to hear educators say the key is teaching students how to prompt an LLM "properly." It doesn't matter how beautiful your prompt is if the "knowledge" base of the tool they're using doesn't actually contain the information you need. Right now, most LLMs are trained to never give a zero result - so if it doesn't have the requested information it just makes up an answer or a source. Even if it has access to a small amount of, say, academic papers, it's not often able to connect those papers to the answer it's produced for the student. Just because it can produce something fast, doesn't mean it's good.

@susannemorr 4 months ago (edited)

I personally don't care to allow spying on myself or other people in order to give AI this sort of data. I don't care if AI cannot operate in the physical world.

@PilsburysDinger 4 months ago

The real problem is “surveillance”

@ggusta1 4 months ago

Scraping reddit didn't work out??? Stunned.

@mattb9664 4 months ago

Keep poisoning that data folks!

@oblonghas 4 months ago

I wanna see it all disappear. It’s not just a dead end, it’s a black hole for humanity.

@alanfarlie759 4 months ago

So the idea is to get the world’s poorest people to train AI so that robots can fold clothes for the rich. What a corrupt and disgusting idea.

@6teeth 4 months ago

Yes, AI did run out of human data. Now they are feeding it synthetic data.

@Kari.F. 4 months ago

I miss the time when you could google the information you needed and actually find it. Now you get an endless stream of information that you don't need and didn't ask for.