0
$\begingroup$

In programming, if a new language could be improved by the language itself, it’s call self-hosting or bootstrapping.

To develop generative AI, there’s some steps, data preparing, model training, fine tuning. Is it possible to use AI it self to help with these steps and make big improvement in efficiency?

$\endgroup$
1
  • 1
    $\begingroup$ Could you clarify what you mean by "AI itself"? That's very broad, it's a bit like asking about using "computers themselves" to improve computers, which is trivially true (chip designs are made on computers). $\endgroup$ Jun 29, 2023 at 6:47

1 Answer 1

1
$\begingroup$

Let me see if this helps-

I believe what you request is an understanding of whether or not Language models generate content to train themselves or other SOTA models.

In short yes, Language models can automate tasks like data pre-processing, and data-preparing, and given the right algorithm (consider OpenAI's new function calling feature) even make decisions on writing the code to train themselves in an optimal manner by mere means of arguing with themselves.

You might be aware of GANs which kind of do that (improve each other without much human interception).

I mean consider all the prompt styles being researched and used, Chain-of-Thought, Tree-based-Chain-of-Thought, etc. If we can have LLM reason itself why can we not have it engineer textual data for us? It is coming, if you search you will find research in this direction. Here is an example.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .