mastodon.world is one of the many independent Mastodon servers you can use to participate in the fediverse.
Generic Mastodon server for anyone to use.

Server stats:

12K
active users

Published a new OpenML.fyi: openml.fyi/2023-08-20/

Big topics:
- tackling the new paper on openness and power from @davidthewid and @Mer__edith
- related, some thoughts on Facebook’s use of “open source” for their LLaMA2 model release
- what would it mean, as a thought experiment, to prohibit open source machine learning?
- and bits and pieces of news catchup. Enjoy!

@luis_in_brief @Mer__edith Appreciate the post, and critique. Looking forward to discussing!

@davidthewid @Mer__edith I told someone today that you were in a difficult position, in that it almost should have been either a shorter blog post or an entire series of books ;) like I said, I have nitpicks, but many of them are hard to cure in a paper-length, time-bounded way. It really is important and glad you published!

@luis_in_brief @davidthewid thanks, Luis! Appreciate your help and engagement.

As a pedant, a small nit: we don't claim that 'open' ONLY concentrates power, simply that in the past it has been leveraged to do so. This is important to emphasize in the current environment where 'open' AI is being conflated with equal access.

I'd also argue that the *possibility of/hope for* future openness does not undermine present claims of closedness. Or, the subjunctive doesn't undermine the indicative.

@Mer__edith @davidthewid 1000% agreed on the second part, and I hoped my endorsement of the conclusion made that clear—but apparently not!

On the first… my reading of the final paper may be colored by my reading of the draft? It definitely seemed like a stronger claim than that to me. But possibly I’m also defensively tainted by the same culture that I point out needs to change (and that I’m glad you’re challenging).

Meredith Whittaker

@luis_in_brief @davidthewid We certainly tried to make the point that what is happening (and what happened in the past) is an instrumentation of open source by corporations in service of increasing power, but that such isn't an inherent quality of open source itself. And that these dynamics are supercharged in the context of "AI".

And again, these are very small nits, and I'm really grateful for your keen eye and willingness to think with us on these issues.

@Mer__edith @davidthewid I may have missed that nuance in the paper (and thoroughly agree).

I’d add that it’s not just a feature of AI/ML; it’s also simply that corporations (including lawyers and VCs) are much more comfortable with the “open” label than they used to be, which removes a layer of … defenses? Capitalism embraces and extends, eventually, as always :/

@luis_in_brief @davidthewid As someone who reported up through Google's OSPO for 11 years this comfort was very much shaped over the long term! (Grateful to have witnessed DiBona and others at the helm of this.) And IMO tracks to the fact that the surveillance business model itself trends toward consolidation, so open "competitors" became less of a threat and more of a business opportunity esp as, in the last 10 years, self-hosting has become a vestige of the past. This and much more!

@Mer__edith @davidthewid agreed—it’s absolutely been shaped, starting with (at least) IBM in the late 1990s/early 2000s paying to educate corporate IP attorneys about how to get comfortable with the GPL. But that’s not been entirely nefarious; lots of people participated out of a genuine sense that open was better for humans, not just the industry.

Updated the newsletter (blog? whatever we call it this week?) by the way: openml.fyi/2023-08-20/

Open(ish) Machine Learning NewsIs this thing on? / 2023-08-20Catching up on power, Facebook, and a range of miscellany.

@Mer__edith @davidthewid yeah, and I’ve never seen anything like it in the open space. Which… is a gap. (Part of why I like your new paper so much is that it is an act of constructive dissensus.)

@luis_in_brief @davidthewid we need a lot more of this, for sure. I touched on the dynamics of capture in AI and particularly its implications for academic research here as well (if you're building a reading list :)

papers.ssrn.com/sol3/papers.cf

@luis_in_brief @Mer__edith @davidthewid ooh I’m very interested in this - but for “open“! would love to discuss more, though Luis might be tired of hearing me rant about structural disincentives to secure the commons 😅

@luis_in_brief @davidthewid oh, I'm not claiming it's nefarious. In OSPO I was surrounded by good hearted people who really believed and IMO did a lot of good.