The measured entropy of the model remains nearly unchanged though which means we have lost capabilities we have not measured, the model hasn't become "denser" it just became more specialized.
replyIt's like comparing two person A and B of similar intelligence where A is smarter and B is a genius at signing, but signing was not on the test so person A won.
Assumes facts not in evidence. Please show your working.
replyhttps://en.wikipedia.org/wiki/Model_collapse - you want to use sigmoid 1.0, but the closer you are to 1.0 the higher the chance your model will collapse so you use 0.99-0.98, but those lead to data loss so after n passes all the original data becomes lost so you have a strict data limit there.
replyThe rest is just the general reality I am sure you are familiar with:
- https://en.wikipedia.org/wiki/Catastrophic_interference
- https://en.wikipedia.org/wiki/Fine-tuning_(deep_learning)
- https://en.wikipedia.org/wiki/Entropy_(information_theory)
You... haven't shown any evidence that we're near collapse. That's what I'm asking you for. Show me some evidence that we are losing capabilities with we have today.
reply