One Thought on the Future of AI Agents World Model

Рет қаралды 2,178

Ай бұрын

A New Vision Language Action Video Mathematics Model? One simple Thought on the Future of AI Agents, regarding its future abilities to include financial or physical simulations in code.
Function calling, Tool use, reasoning in LLMs, Planning AI, Science AI.
A Vision Language Action Video Mathematics Model as the new WORLD MODEL for AI?
#airesearch
#ainews

Пікірлер: 17

@richardchinnis Ай бұрын

This is what has been rattling around in my head for a long time. And I think it’s similar to what LeCun is saying. What we have is a model that produces language. Our language is not our intelligence. It is the communication of intelligence. It’s the final product. The wrapping around the true intelligence. The LLM should be used to encode my communication to the agent, then the intelligence (not LLM) goes to work, then the result is communicated back to me using an LLM. What we have at the moment is a very convincing and entertaining flea circus.

@agsystems8220 Ай бұрын

"Our Language is not our intelligence". I would dispute that. Where human intelligence exceeds animal intelligence is in abstract tasks, and abstract planning. (conjecture on my part from here) At it's core we are able to do this and meme-ify this because we have this tool at our disposal for communicating non existent things in a concrete way. We are then able to ponder on them, and develop structures in our brain that can accelerate this reasoning. Eventually the majority of our intelligence will live in these acceleration structures, but they were built from the integration of language. The core surprise about LLMs was that they did seem to be doing some reasoning. There are at least a couple of real fleas in this circus. I actually think we work similarly, and we are seeing a spark of intelligence similar to ours. Getting from a spark to a fire is not trivial though. I don't think the future will be monolithic LLMs. I think we can do much better with good architecture, especially around tool use, but I do believe that language will be very deeply rooted into any genuine intelligence we create.

@LamontCranston-qh2rv Ай бұрын

Genius! Computational Bio-chemistry, for example, has been around for decades, but it typically uses a "brute force" approach: trying every possible combination. An "intelligent" version of that would be exponentially more efficient no doubt!

@AGIBreakout Ай бұрын

👍Check out the Playlist Physics Informed Machine Learning❗❗

@LamontCranston-qh2rv Ай бұрын

In order to do math, deep learning models would need some sort of spatial reasoning. For example, the summation symbol. It has spatial references. The superscript, if any, the subscript, if any, what follows to the right, and whatever precedes to the left, if anything. Mathematical symbology is spatially referenced. Even simple division has a "top" (numerator) and a "bottom" (denominator.)

@ftlcpatent9942 Ай бұрын

Nice I like it. I have been working on combining 31 model types in to one new model and your work is reinforcing and exciting. Thank you I hope more people see and understand your it's Fresh and cutting Edge.

@MecchaKakkoi Ай бұрын

Agentic modularity also has some security benefits

@johnkintree763 Ай бұрын

My preferred architecture is open source and decentralized function calling language models running on personal devices such as smartphones that collaborate in building a shared world model that can include links to repositories of specialized models such as fluid dynamics, climate change, economics, and so on.

@jichaelmorgan3796 Ай бұрын

Eventually distrubuted intelligence across the internet

@attilalukacs1981 Ай бұрын

What if this Mathematical AI System is already here in the form of the Wolfram Language? I know it is not fully that system that you described in the video, but integrating it with an LLM can be a good starting point.

@bastabey2652 Ай бұрын

AI Agent = LLM + $$$$$$$$$$$ with the existing prices of high end LLM, the agentic AI is a roadmap to more money to AI companies

@pedromoya9127 Ай бұрын

great video, perhaps Wolfram's TOE computational paradigm can be used and boosted with the development science-physical model

@thesimplicitylifestyle Ай бұрын

😎🤖

@NeuroScientician Ай бұрын

Isn't LLM more of high school geometry/trig that language?

@ArianeQube Ай бұрын

No, it's linear algebra and matrices. No geometry and trig in LLMs

@hotbit7327 Ай бұрын

More advanced than high school level, but in essence, yes, it's all languages. In geometry you have - nouns (point, line) - verbs (rotation, translation) Using already known words you create new words according to the geometry language. N x N matrix can be interpreted as a N dimensional number, at least when N = 2^k. For example 2x2 matrix and notation a+ib both can be interpreted as a complex number.