Archive

Open is dead! Open is everywhere! Long live open, long live the *praxis* of open. Which is what I'm going to go do some more of for a while.

If the first wave of LLM criticism is on life-support, what ethical questions do we face now? And how can/should open engage?

Thinking about actually writing code instead of writing newsletters, as well as a lot of other open+ml news.

Links on culture, tools, and SAP in a trenchcoat.

Vocabulary, ethics, and disclosures, and some news and links from The Week That Was in AI + Open.

About the next phase of the newsletter—what happens to open source when the LLM is an agent of change?

In which I both celebrate and mourn the data commons.

Getting angry about Senator Vance’s call for… open? and some good examples of what smart open could look like.

Highlights from recent news, including a few interesting new true-open models but lots of negative signs from other quarters of the open/AI culture wars.

Responsible licensing and non-consensual pornography; new open-ish Pile coming; and how openml.fyi will evolve in 2024.

Many small notes, plus three long-simmering micro-essays—two on OpenAI and Facebook, and one on how the switch from public licenses to 1:1 contracts might impact open.

Lots of links on the state of regulation, the recent transparency report, and more.

So much news, so little time.

Catching up on power, Facebook, and a range of miscellany.

I’ve been writing this newsletter for about six months, so I thought it might be a good time to pause the news firehose, and instead review and synthesize what I’ve learned about the potential for open(ish?) machine learning. That took a while — so thanks for your patience, and welcome back!

Micro-essays on the commons and model copyrightability; new open(ish) tools; and a lot more.

Upcoming public panel moderation; shifting power via foundation models (open or not); public ability to understand via great WaPo visualization; modularity; and more.

A busy week: copyrightability, throwbacks to late-90s monopoly concerns, who is a coder anyway, permission culture revisited, artistic typography, and so much more.

Two mini-essays on the Open Data Definition and why OpenAI’s non-profit/for-profit split can be legal (but might not be), and many news bytes from a busy week.

Focusing on some key themes, via deployment on laptops, government algorithms, a blog post from Anthropic, an intro to foundation models, and more.

An observation from the Linux Foundation Legal Summit; Apache grapples with ML; adversarial data; memes explained; and muuuch more.

Long pondering on when the giants of the space might fight for fair use, or compromise; an observation on the state of open data governance; news on attribution; memorization of (some images); and two weeks of miscellany.

Updates, including data from Hugging Face, new litigation, and thoughts on power.

What I’m *not* doing in 2023 (predicting the future), a few examples of this tech getting simpler and faster, and an extensive discussion of Wikipedia.

I've been under the weather and busy at work, so things have been sporadic. I look forward to doing this more steadily in the new year. The rest of the world has not slowed down... Holiday ML poetry Every child there’s been From Aaron to Zeta We can build an AI Because we have the data Read the whole delightful thing, from Casey Fiesler. Top-down 'innovation' This preliminary report on data licensing for AI from the Global Partnership on Artificial Intelligence is, at some level, a pretty

Avoiding ChatGPT; updates on economics (trends going in good directions), law (it’s messy), explainability (nothing good).

RAIL adoption; observations on “release early, release often” in ML; and a micro-essay on transparency.

The newsletter has been quiet, but the world has not. Some thoughts after a couple of very busy weeks.

No big meta-themes this week. Enjoy Halloween :)

Litigation news, model copyright, stable-diffusion 1.5 delayed (and maybe taken down?), transparency, auditability, and a few events.

An experiment in re-organizing the newsletter, based on a list of the values, techniques, and joys I value in traditional open.

Notes on new government initiatives, how ml is reminding me of late-90s open, open data, and more.

Highlights: discussion of proper terminology (key if we’re going to get this right); examples of the network effects of open; the possible end of NO WARRANTY; an interesting academic paper; and some fun/scary demos.

My second email to friends on open+ML. Archived here with only light edits.

My first email to some friends on this topic; the positive response to this is part of why I decided to start the newsletter.