Thinking is not a linear straight phenomen. Let' create a corpus on real human thinking for the initial learning of an LLM

ChatGPT Feature requests
1

The essential is in the topic
LLM are trained on polished texts, where all reflexions seems to be nice demonstrations, from A to B.

Real life thinking is more chaotic, with try, failures, fool ideas…

Real life thinking can say I don’t know. Can say maybe. Can say no to established knowlege. Doubt. Can also see the difference betwin trust and faith.

I propose to the community a project to build a corpus of real “savage natural” pieces of thinking.

Try to do initial leaening of an LLM with this corpus
Fine tune after with specialized knowledge.

One of the goal is to restrict hallucinations

2

Finaly, my idea is maybe pure bullshit, I have no answer, even to criticize, ridiculize my naivety.
Or maybe I am nearly invisible as a new member…

3

I am not here to hear people saying my idea is wonderfull.

I 'm hoping hard critical analysis.

Why my idea is bad, or impossible.
And, maybe, after that.
How to make it less impossible.

What I call dinosaurus models ( larger and larger) is for me an evolutive deadlock. Models should be smaller and agile. Efficiency is the key, even if at first we lose effectiveness.