Right, I mentioned that the more they’re used the more incentive there is to train new ones. But also it’s not like buying a product from the store which then creates a demand for a replacement, it’s much fuzzier and it’s difficult to point to any one user as increasing the demand for more models. Realistically I think we’ve gone well past the point where it makes sense to train general purpose LLMs any further, it’s drastically diminishing returns for marginally higher quality. The continued fervor is being driven by the same people that drove crypto and NFT prices way beyond reality: speculators and VC parasites trying to shove the new hype buzzword into anything they can get their hands on.
Right, I mentioned that the more they’re used the more incentive there is to train new ones. But also it’s not like buying a product from the store which then creates a demand for a replacement, it’s much fuzzier and it’s difficult to point to any one user as increasing the demand for more models. Realistically I think we’ve gone well past the point where it makes sense to train general purpose LLMs any further, it’s drastically diminishing returns for marginally higher quality. The continued fervor is being driven by the same people that drove crypto and NFT prices way beyond reality: speculators and VC parasites trying to shove the new hype buzzword into anything they can get their hands on.