You can still have a conference for AI safety specifically and present at both conferences, with a caveat. From NeurIPS: > Can I submit work that is in submission to, has been accepted to, or has been published in a non-archival venue (e.g. arXiv or a workshop without any official proceedings)?Answer: Yes, as long as this does not violate the other venue’s policy on dual submissions (if it has one).
The AI Safety conference couldn’t have an official proceeding. This would still be great for networking and disseminating ideas, which is definitely worth it.
Another option may be a conference (that forms more of a Schelling point in the field than all the existing ones). These seem to be more popular in the wider field. But both solutions also have the risk that fewer people outside AI safety may read the AI safety papers.
Academic AI Safety Journal
Start an Academic Journal for AI Safety Research
Problem: There isn’t one. There should be. It would boost prestige and attract more talent to the field.
Solution: Fund someone to start one.
This has come up a few times before and is controversial.
Pros:
more incentive for academics to work on pure safety without shoehorning their work
higher status
better peer review / less groupthink
Cons:
risks putting safety into an isolated ghetto. Currently a lot of safety stuff is published in the best conferences
Journals matter 100x less than conferences in ML
I think academics are a minority in AIS at the moment (weighted by my subjective sense of importance anyway)
FWIW I take the first con to be decisive against it. Higher status takes a long time to build, and better peer review is (sadly) a mirage.
You can still have a conference for AI safety specifically and present at both conferences, with a caveat. From NeurIPS:
> Can I submit work that is in submission to, has been accepted to, or has been published in a non-archival venue (e.g. arXiv or a workshop without any official proceedings)? Answer: Yes, as long as this does not violate the other venue’s policy on dual submissions (if it has one).
The AI Safety conference couldn’t have an official proceeding. This would still be great for networking and disseminating ideas, which is definitely worth it.
Another option may be a conference (that forms more of a Schelling point in the field than all the existing ones). These seem to be more popular in the wider field. But both solutions also have the risk that fewer people outside AI safety may read the AI safety papers.