Thanks! Yes, there are many similar things. But the ChatGPT seems to have the most amazing performance.

It's a sad story as I put a lot of time on generation during these years. Any possible suggestions that our research can focus on?

Yes, it is quite costy. However, it seems not easy to modify it in our research as it is not open.

But google really change the way we are working. This is why I guess there may be another change.

Seems recently, not too much paper are doing on them. Don't look at details. Maybe models like OPT is still too large?

I guess openai will not open the model for us to apply a meta layer. It will remain a black box. So, this is why we cannot do anything on top of it.

Yes. That's the benefit of in the big companies. However, for a lot of NLP researchers like me, we do not have that many gpu resources(I believe most of the companies also cannot afford this).

I have tried many cases. For example. It gives correct proof of one of my technical lemmas in my own paper which make me quite amazine. It is a simple lemma, but it is very specific to my question. I also tried to search with google but do not find the answer.

I have tried and found it is a huge advance in this area. It not just googling. It has some inference capability.

It's not just googling. I can summarize the information it has and write a good answer to the questions. It can even have some inference capability.

I got a strong rejection with only a few sentences of weakness. Never seen this before. Seems the reviewer gets extremely angry.