Thanks for another awesome thread, the results are surprising and not.
My big hope right now is that openai has some pride and improves their base offerings. As of right now, the majority using their services would be served as well or better by opensource or ‘open-kinda source’ models.
I would imagine the branding of “industry leader” in ai to “their main service is bad” has to hurt at a pride level. Yes gpt-4 is fucking amazing. But no body cares about your handsome son off in college if you introduce them to “jerry with the pasta sauce on his shirt” first.
Personally, I hope all you sexy-brained creators drink their moat and use the marshland for rice.
Thanks again for the time and hard-work you put into testing, investigating and sharing your results. You and chat arena are the two most reliable benchmarks I have. Cheers and may good fortune find ya.
I agree with the sentiment. A contribution is a contribution. Even an utterly failed contribution can have wonderful results when shared.
If, my critiques of the writing quality and content of certain recent papers played a part in this post, I apologize. In my case when I say improvements in ‘transformer kung-fu’ it is my attempt to advertise “we have now exceeded this humans education and understanding”.
All that said. I think some papers are of such a poor and offensive quality, that getting publicly dunked on for it, is a great way to keep liars from stealing grant money/funding from people doing real work. I am already annoyed that when a member from here contributes something, they don’t mention that human, they mention the community. There is no reason a valid contribution shared here, incorporated into an academic or corporate project shouldn’t and couldn’t be credited.
So, I agree with your points and also am probably guilty of the same of the type of comments you speaking out against.
Just another thing I can improve on. Cheers.