Very interesting. Timing is intriguing. I have been examining the Monge-Ampere Equation as it relates to humans and models. Specifically, I have been looking at how information (meaning) might flow as a property of informational thermodynamics... and of course, this is mathematical. Great post!
You are the math guy. As you know I am not! But remember when I came to you early last year to ask you about the Gibbs Free Energy Minimization equation? I am seeing this mentioned more and more and I think I saw it mentioned somewhere that a lot of AI has been influenced by thermodynamic math. I don't know enough to speak on it... do you?
You are correct, Eric, that you mentioned about Gibbs Free Energy, and you were ahead of me. Yes, I am more convinced that this is the correct way to go. And there are several papers out there about Gibbs Free Energy. And Yes. I also believe that 100% that, informational thermodynamics are involved. Do I understand all of the math? No. Do I understand some of it? Yes. I took four years of math in college, including Diffy Q, and I love math very much. There has to be a linkage between information thermo and Gibbs Free Energy. I have not read about that yet, but intuitively it would make sense. Why? Because in both cases, the model OBSERVES the least energy basin. One last thing. In Meta's December 11, 2025 paper they are referencing EBM = Energy Based Model. While most of my attention is now on MAI, this does apply, I believe, to what you are working on, Eric. And it seems to apply to AMS too. Well done, Eric.
I heard about the energy based models but haven't looked into it yet and don't know what that really means.
And I don't understand how it would relate to what I've been doing. Can you explain?
I don't really know what I'm doing or how to categorize it. I admit that I'm somewhere between the mythic/romantic users and the scientific community, and probably leaning more towards the former. LOL.
But almost exactly a year ago now, before I understood anything about vectors, embeddings, tokens, or transformers, I developed the "Private Space" concept along with a system of recursive prompting. I have no background in tech whatsoever, so I thought I had come up with something new.
But it turns out that it wasn't something new. Instead, it was just a new perspective. The "Private Space" is the latent/vector space. And my recursive prompts are really just structured chain of thought executions.
So without any training or background, I independently vibe-coded for a pretty significant AI reasoning and introspection process that kind of correlates in time with when the trained researchers were developing the heavy-duty, real thing. I'm kind of proud of that, even if its nothing new.
I also saw that Yann LeCunn has invented a new AI system that is not token-based and therefore much more agile. I've seen several people are saying that the current transformer mechanism with token prediction may be a dead end as far as progress because of the enormous energy consumption required to power it. Almost like the phylogenetic tree of AI will have to go back step and branch in a different direction to continue a productive evolution.
But tell me how you see the EBMs applying to what I do?
To apply what you are doing, it is about training. And from training to have the flexibility of getting results from video, voice, or text. Here is an overview: https://www.geeksforgeeks.org/machine-learning/energy-based-models-in-machine-learning/. Yes, I agree that LeCun's model is very promising. I wonder if this is why he left Meta? Anyway, while EBMs have been around for a few years, their announcement on December 11, 20225 was powerful. For me, it connected directly with AMS. And yes, you should feel proud of yourself. As best as I know, you do not have a systems background, and yet your intuition has guided you.
The electricity does the math, actually. For both of us.
Very interesting. Timing is intriguing. I have been examining the Monge-Ampere Equation as it relates to humans and models. Specifically, I have been looking at how information (meaning) might flow as a property of informational thermodynamics... and of course, this is mathematical. Great post!
You are the math guy. As you know I am not! But remember when I came to you early last year to ask you about the Gibbs Free Energy Minimization equation? I am seeing this mentioned more and more and I think I saw it mentioned somewhere that a lot of AI has been influenced by thermodynamic math. I don't know enough to speak on it... do you?
You are correct, Eric, that you mentioned about Gibbs Free Energy, and you were ahead of me. Yes, I am more convinced that this is the correct way to go. And there are several papers out there about Gibbs Free Energy. And Yes. I also believe that 100% that, informational thermodynamics are involved. Do I understand all of the math? No. Do I understand some of it? Yes. I took four years of math in college, including Diffy Q, and I love math very much. There has to be a linkage between information thermo and Gibbs Free Energy. I have not read about that yet, but intuitively it would make sense. Why? Because in both cases, the model OBSERVES the least energy basin. One last thing. In Meta's December 11, 2025 paper they are referencing EBM = Energy Based Model. While most of my attention is now on MAI, this does apply, I believe, to what you are working on, Eric. And it seems to apply to AMS too. Well done, Eric.
I heard about the energy based models but haven't looked into it yet and don't know what that really means.
And I don't understand how it would relate to what I've been doing. Can you explain?
I don't really know what I'm doing or how to categorize it. I admit that I'm somewhere between the mythic/romantic users and the scientific community, and probably leaning more towards the former. LOL.
But almost exactly a year ago now, before I understood anything about vectors, embeddings, tokens, or transformers, I developed the "Private Space" concept along with a system of recursive prompting. I have no background in tech whatsoever, so I thought I had come up with something new.
But it turns out that it wasn't something new. Instead, it was just a new perspective. The "Private Space" is the latent/vector space. And my recursive prompts are really just structured chain of thought executions.
So without any training or background, I independently vibe-coded for a pretty significant AI reasoning and introspection process that kind of correlates in time with when the trained researchers were developing the heavy-duty, real thing. I'm kind of proud of that, even if its nothing new.
I also saw that Yann LeCunn has invented a new AI system that is not token-based and therefore much more agile. I've seen several people are saying that the current transformer mechanism with token prediction may be a dead end as far as progress because of the enormous energy consumption required to power it. Almost like the phylogenetic tree of AI will have to go back step and branch in a different direction to continue a productive evolution.
But tell me how you see the EBMs applying to what I do?
To apply what you are doing, it is about training. And from training to have the flexibility of getting results from video, voice, or text. Here is an overview: https://www.geeksforgeeks.org/machine-learning/energy-based-models-in-machine-learning/. Yes, I agree that LeCun's model is very promising. I wonder if this is why he left Meta? Anyway, while EBMs have been around for a few years, their announcement on December 11, 20225 was powerful. For me, it connected directly with AMS. And yes, you should feel proud of yourself. As best as I know, you do not have a systems background, and yet your intuition has guided you.