Language through knowledge vs. knowledge through language

ChatGPT works like an interactive encyclopedia, remembering standard responses, questions, and statements. But, if you add enough data, and if it learns enough from interacting with users, it can become as intelligent as a human being. This is the “language through knowledge” approach of AI.

Before I knew about ChatGPT, I devised a method to create an intelligent AI. Have many, many programmers teach it English, each programmer teaching it certain words and their associations, then give it information and have it interact with people to increase its knowledge and understanding. This is the “knowledge through language” approach.

Hi welcome to the community! Not sure what you’re trying to accomplish with this post (a discussion or debate or?) but it might be best to share some links to support your theory or any more information you have to clarify. As of now, chatGPT is not an encyclopedia. You might want to do some research on how LLMs work. I’ve never heard mention of “language through knowledge” in any AI related subject.

1 Like

It just means that ChatGPT uses a library of standard answers, questions, and statements to function. This is my understanding of ChatGPT. As it learns more answers, statements and questions it will eventually be able to think with the same intelligence as a human being. This is what I mean by “language through knowledge.” It goes from a limited repertoire of data to full understanding of language, by collecting many, many examples of answers/questions/statements. Each person talking to it has a unique question/statement, so by encountering many unique questions/statements, it will eventually learn to answer any question. Just like a student learns to answer any math problem by tackling specific problems.

I have been thinking what AI can be used for, with its great processing power. I have several ideas:

  1. Creating and running intelligent video games that adapt to the user(s) as if the NPCs and game world were real.

  2. Designing intelligent machines.

  3. Designing cybernetics.

  4. Analyzing the human brain, in general or individuals’ brains.

  5. Analyzing and editing genes. This has the application of putting evolution into our own hands. An example of an evolution: give a man an ovary and a woman a testicle. Then, they can “get pregnant with themselves” and the embryo can replace their tissues to renew their bodies. That way they could live across generations, for perhaps hundreds of millions of years, until true immortality is developed, then they could live forever. Or, alternatively, you could edit genes of stem cells to accomplish the same goal, or just edit any kind of cell to replace tissues.

I am sure there are other applications, but I can’t think of any at the moment.

Hi, these are nice musings, but are you trying to have a conversation here? Please try and post things that are functionally useful in the form of debate, questions, or discussions. You might want to post things like this on Reddit etc.


Well, for example, other people can share their ideas for AI applications. Communicating these ideas is important to make them a reality. One day ChatGPT will be capable of doing these things. I don’t think I’m doing anything wrong.

I wonder if we could combine “language through knowledge” and “knowledge through language…” What about having ChatGPT practice all kinds of sentences to become fluent in English, like a Babbel course, with image associations and other associations of the words, then have it read, for example, novels, like Great Expectations, then write an essay about them? Then we critique the essay, so it learns to read and write better.

And as for the “AI applications” thing, as opposed to adding opposite sex sex organs, we could create an organ designed to periodically renew tissues in a way that would be as sustainable as reproduction, thereby achieving the same result. What I mean by “as sustainable as reproduction” is that we could live as long as the species could potentially live across generations.

Why not have ChatGPT search google for specific words, then an operator indicates the images/definitions that actually fit the word?

Another thing we can do is have “chunking” like in the human brain, like small groups of associations, larger groups, etc… a mind organization system.

Combining those two, you pack all associations in their appropriate places in the AI’s “mind.”

Oh, and also ChatGPT can start searching sentences on google, and an operator can tell it which images/results are accurate. To build its understanding.