Many people these days are shocked by the OpenAI announcements on results on open (until recently) problems and they ask the company to stop testing frontier models on open math problems. That doesn't sound right. Just because you cannot compete technology you cannot ask it to cease to exist.
Nevertheless, it is reasonable to ask AI companies to make such announcements in a slower pace. Not necessarily to slow down model development, just to make such releases on smaller packages and less frequently, so that the community will have time to digest the results and perhaps even understand the proves.
Bombarding the community with a huge bundle of information can only make the whole task unmanageable. That is if you want to make some serious contribution to scientific progress and not just make good impression to your stockholders.
No comments:
Post a Comment