♾️ Tranfer off of openai to open source model (Possibly Vicuna/FastChat)

[pmb2]

Summary 💡

I would love to see us get away from being dependent on openai's model.
We should train our own.

Examples 🌈

https://github.com/lm-sys/FastChat

Motivation 🔦

If they start tainting the models or take them from us..
No sense in waiting, we can make our own.

[russinet]

I think you severely underestimate the amount of sheer processing power and time required to train a model comparable to GPT-4, not to mention actually running it. There are a couple models such as GPT-J, Bloom, or LLama that you might want to take a look at, but don't expect anything near the quality of GPT-4.

[nponeccop]

The cross-model capability and/or ability to use weaker, more open and/or more tunable models as agents for GPT-4 would still be a good capability to have though. Delegate more, use GPT-4 for only the hardest tasks.

Should we close it as a duplicate of #438 and #414 ?

[erkkimon]

Also Open Assistant develops fast. I guess there will also be decentralized computing as an option soon. If the open locally run language models are not good enough today, they will be tomorrow.

[pmb2]

Yeah i just read about open assistant, I think we're doing better here in alot of areas, but there is alot to learn.
Things are moving quickly.

[Qoyyuum]

Closing this issue in favor of and duplicate to #567