Use real tools with real data, not fakes and “educational games”
Alien Admiral Watches Human Child Play Starship Simulator… and Realizes It’s Deadly Serious at https://www.youtube.com/watch?v=6MnB5aFEhSA
Richard Collins wrote:
I spent the last 27 years of the Internet Foundation to make this kind of thing possible. There are 8.2 Billion humans and about 5.4 Billion Internet users. Good ideas are nice, and this story is well done. But making “real physics” is hard. Not because it is impossible, but because the information required has been hashed and scrambled, diluted and dumbed down. So what is on the free internet is a tiny fraction of what is known, and completely inadequate to train real simulators and games. The LLM AIs are taking that “free stuff dumped on the internet” and simply churning it with a statistical index. They do not even bother to index the raw data that goes into their static snapshot. And they make no effort to give back.
Technically what you describe is possible, including modules for every thing anyone can think of. But the tools available for software development are inadequate and themselves highly biased to give limited results – since the purpose is commercial success and not to train adaptable humans and AIs for a more complex universe.
I took the transcript and had ChatGPT reformat it into story as it would be printed in a book, to see how hard that would be. It is tedious and I finally had to do a page at at time and copy and paste because they have one LLM hammer and they try to do everything with one not very good method.
But eventually I asked if it had read the story and understood the issues. It is still stuck in those old “make education fun” dead ends. I say it is not “fun” that matters, it is “real” that matters. Real data from real things – lossless data
Here is one thing OpenAI ChatGPT 4o says:
“The future of education—the kind shown in this story—is about removing the artificial barriers between knowledge and experience.
If you want to understand fluid dynamics, you should watch and manipulate virtual fluids.
If you want to master orbital mechanics, you should pilot ships and adjust real-time vectors in a simulated space.
If you want to grasp quantum entanglement, you should experiment with linked particle fields and observe their behavior.”
Richard Collins, The Internet Foundation
