Is it possible to build a model on top of a model?

Research
1

When I see the bio models developed for medical research, only one thing comes to mind: why isn’t a model built on top of these bio models?
When I think of metaphors like a cell repairing itself, a cell replicating itself, approaching or moving away to interact, or feeding, I wonder whether we could turn an LLM into an internally self-improving structure to enhance its reasoning. I’m curious if we could leverage the movement and cellular characteristics of a biological organism to make the neurons of text-based models more functional within this framework.

1 Like
2

Well, I do not know very much but I did see some conversation about “echoing” where output is post-processed.
How are you?

3

I’m only aware of one thing: we don’t use the parameters of existing models to improve their own reasoning.

4

My friend, you are ahead of me in school.

I am yet to get my workstation put together and to run AI.
Anything I have is conceptual and conversational. I simply want to talk to the folks, of this AI age, here on Hugging Face.

If a function produces a language that other functions can understand then the sky is the limit for AI, in my opinion.

I vote for a universal structure to it all.

5

Mostly I’d guess it would be down to refinements then retraining as even a small change on pre loaded weights can have undesirable effects , when layering ontop of models , it’s easiest to keep it to modules that can snap in and out with enough compatible modules it could almost be a model on top of a model :thinking:
For the organic self repairs and growth You might want to Google search madmoo-Pi then the git safety conscious ai I get flagged if I post links :grin::vulcan_salute:

6

Appreciate your Idea that make AI copy how cells work - self-repair, grow, and get smarter on their own, but Scientists are building AI that can change itself while working, just like you described.

You can meet, implement and discuss with experts this plan

  1. Use existing “self-learning” AI as the base
  2. Add cell-like behaviors (connecting, growing, improving)
  3. Start small and test it.
1 Like
7

Welcome to HF @Angelina067