Can open-source models replace ChatGPT?

I believe that while they have the potential to be powerful tools, they cannot completely replace ChatGPT's ability to generate natural and coherent responses in a wide range of contexts.

If someone or a crowd can fund an OS project for millions of dollars just to train models, why not.

Certainly, open-source models can do so. But my twist here is that they should be allowed to flourish in the market, and large enterprises who may have their own agenda should refrain from colluding. But on a product capability basis? Sure!

We seem to be moving in that direction, but not quite there yet. Current open-source models can match the capabilities till GPT-3 or GPT-3.5, but it's still far off from GPT-4 / ChatGPT. But this has been the trend so far, open source models take time and usually lag behind the commercial ones, till they hit an inflexion point, and actually start improving and beating closed-source ones. We hope that the inflexion point comes soon. Another direction would be that we get smaller, use-case-specific models which beat ChatGPT, trained on your own data or data that you're interested in.

I think it is possible.

I don't think so, ChatGPT is gonna get even stronger and more powerful in terms of performance. So eventually people will stick to ChatGPT..

From a business perspective, the pricing of ChatGPT is hard to beat. If you want to self-host, or host in a cloud, you are going to need some serious scale for that investment to be worthwhile.