3 ms·
As a general principle, when there is a "they-said she-said" situation, and the certain facts about one side of the story don't add up, from my perspective that
by oh_sigh 6y ago
As a general principle, when there is a "they-said she-said" situation, and the certain facts about one side of the story don't add up, from my perspective that increases the odds that the other side is telling the truth.
I have no idea what Google's standards are for letting a paper be published. My guess is that they don't make standards up on the fly, and no unique standards are applied to Timnit and not to literally everyone else.
> Keep in mind the feedback was that you cannot publish this paper, and that isn't disputed by Google.
Yes, and that in itself makes me more likely to think there is some benign, standard reason for not allowing publication rather than some nefarious motive about suppressing Timnit or hiding their own guilt, or whatever reason Timnit has provided as to why she thinks Google doesn't want the paper published.
If the argument is "Google won't let me publish this paper because it's too dangerous for them", and then I look and see that there isn't really anything dangerous for Google in the paper, then I would say that perhaps the argument is incorrect as to the reasons Google wouldn't let them publish
- joshuamorton 6y ago> My guess is that they don't make standards up on the fly, and no unique standards are applied to Timnit and not to literally everyone else. It is not disputed by Google that special standards (an additional, non-standard review process) were applied to this paper. There is some lack of clarity about how often that additional review process is applied (https://artificialintelligence-news.com/2020/12/24/google-telling-scientists-give-ai-positive-spin/ https://artificialintelligence-news.com/2020/12/24/google-te..., https://www.reuters.com/article/us-alphabet-google-research-exclusive-idUSKBN2AP1AC https://www.reuters.com/article/us-alphabet-google-research-...). But broadly it seems to not have much to do with the technical merit of the papers, despite what Google originally claimed, and instead be a legal/PR process.
- cjbenedikt 6y ago...or you can argue exactly the opposite along the same lines...
- oh_sigh 6y agoWhat? If someone says a paper is too dangerous for Google, but then I read the paper and there is nothing dangerous at all for Google - that is evidence that there is in fact something too dangerous for Google in the paper?
- joshuamorton 6y agoI think the argument is moreso that Google won't even allow milquetoast criticism of LLMs, which falls perfectly in line with the events.
- majormajor 6y agoIf Google had a big problem with something you find mild, that could easily make you believe Google is making errors in judgment.