The philosophical comparison is simulation versus duplication. Today, travel for example is much easier to simulate than duplicate. In a simulation we easily travel between stars but we cannot start to duplicate that travel. While competition is focused on today’s limited artificial neural networks the limitations exclude duplication.

The ultimate question about today's AI is why it gets compared to an intelligent being when it is made of human defined statistics. And then it is said to cover ALL use cases! That is the ultimate goal of strong AI defined in the 1950s that remains unsolved despite the claims. By focussing on domains that work with statistical models and that can tolerate errors, applications that have value can be developed. But I guess the value of that subset is too limited to create high-value companies.

The alternative model is that the money lost on endless training of AI systems that don’t provide value and are loss making will go back to models that are profitable for very narrow use cases. Today’s AI will be a warning to others not following scientist principles. The hype around systems that are fictional will go the way of other false starts to scientific models.

"human value is not just accuracy." And worse, today's AI is not accurate, anyway. Perhaps that's why Wozniak is disappointed? And at the same time, other CEOs from Silicon Valley are claiming that this year, agents will show amazing capabilities. After 3 years of this "AI-revolution," perhaps it is time to just stop all the predictions. They are not coming true anyway. Instead, let's see what happens and talk about what they can do, not predictions of what they might do in the 1-3 year future.

Again critical thinking is useful in new ventures. How does a statistical text generator solve science problems? Dos anyone other than ai ceos believe that science is the manipulation of word sequences or instead the interactions of real world entities? Describing gravity in words is less powerful than newton’s f=ma because the equation tells a very clear story of motion at low speeds compared to light.

Let me fix this: "We used to retrieve information. Now we generate it in real time." Fixed: "We used to retrieve information. Now we generate HALLUCINATIONS in real time." AND "systems that think in context." Fixed: "Systems that GENERATE STATISTICAL TEXT BASED ON LOSSY REPRESENTATIONS." LLMs don't THINK. They don't use human-like context.

Why are LLMs AGI (not). The historical figure is Julius Caesar, but people would confirm that's what it means, first. The LLM just choses the probable, and wrong in this case, possibility. Note in the quote, he said it yesterday. That means, where I am, Tuesday. The text gets the day wrong from a simple English sentence. The translation of "he came,..." is not the Latin from the historical figure (I came, ...). So there are similarities, but terrible errors that humans don't make. The errors in recognizing current context is one of the failures of LLMs currently a design flaw with that kind of 'AI.'

This is the link to the full article: substack.com/@johnatpat/note…

Science is most effective when given a reason for your thoughts. "That branch of AI is lethal!" If it were true, then there would need to be discussion. Will intelligent machines, once they can move and talk like us, be a danger? Why would a machine that is like us become a danger, unless people give them the ability to make decisions without supervision?

Not entirely correct @ChombaBupe . Cognitive scientist @jbthinking has a working implementation for Language understanding on a machine from his brain theory. We agree today’s generative AI has more to do with dangling carrots than brain emulation.

AI Improves with Cognitive Science - Part 2 This incorporates the analysis of the science of brains to show why AI struggles with brain-like function: Moravec's paradox. The answer to how the brain processes information is answered: it doesn't. A different model that distributes a representation of the world is explained in steps. The next article can then use this to explain how human language performs some of the impressive tasks of generalization and understanding. The issue to solve comes from how computers work. Encoding data and processing it with a program is alien to what a brain does. How would a binary code be created to represent vision or hearing, for example? Brains are more capable than our best machines today. We talk, move around and use languages that seem to be coded in a way our best scientists have failed to emulate. But that just comes back to how our brain works. Brains aren't based on mathematical models. Today's article will be a chapter in my upcoming book, but for now as is the goals of science, please read the article and criticise its ideas if you can. I expect it will be hard to criticise because the model has been used to build a working parser that operates in real time, on a computer, that uses the model as described. It converts the meaning of words and phrases into related representations aligned with the world's languages (refer to the RRG linguistic model). The article explains what a pattern is, how patterns can be manipulated with exquisite detail as we see in human interaction aacross all our sensory modalities, how motor control works with the same brain material, and how this model can be extended to create language. The hard work comes from the animal brains that evolved to human brains. When we refer to someone's face, you can almost see them and it doesn't get confused between other people's. What magic does that without data or labels. It's automatic and what brains do!

It seems strange to take what animals do from very early life and seek to learn with simulations instead. It’s better to accept that machine learning is the wrong approach and therefore seeking data to train on it is unnecessary. There has been a lot of effort to get ML to handle robotics but the better approach is to use science and create better models.

Me: How does one solve spatial intelligence? Fei-Fei Li: By creating 3D world simulations. Me: I see. Well, good luck with that. 🙄 PS. It's a good thing babies don't need a 3D computer simulation to gain spatial intelligence, isn't it? The real world is free and perfect. 😀