Sign in

Aviya Skowron

@aviya.bsky.social
844 followers 404 following 88 posts

Head of Policy at EleutherAI. They/them. Working at the intersection of open source and AI policy. Former philosophlete.

PostsRepliesMedia
Aviya Skowron @aviya.bsky.social · 19/09/2026
reasonable take: METR has some conflicts of interest with Anthropic, raising questions about whether they can evaluate truly independently monkey's paw:
000
Reposted by Aviya Skowron
Raffi Melkonian @rmfifthcircuit.bsky.social · 06/07/2026
Every soccer match should be followed by appeals of each foul. That is the only way soccer can become more like appellate litigation, which I understand to be the goal of all.
361237180
Reposted by Aviya Skowron
joannes @playthroughline.com · 30/06/2026
A vicious cycle of four steps surrounding a black cat with a thousand-yard stare and a dapper little tie. The four endlessly cycling steps read:

-going forward we're gonna need more of your data to make sure it's you

-ahaha youre not gonna believe this but we had a bit of a data breach

-your data is probably for sale online now

-that means someone could easily impersonate you
23105634518
Aviya Skowron @aviya.bsky.social · 05/06/2026
Some state legislatures that have figured it out, and yet!
030
Aviya Skowron @aviya.bsky.social · 05/06/2026
love the feeling of getting more frustrated the longer you try to work on something
050
Aviya Skowron @aviya.bsky.social · 09/04/2026
I'd also read this (much shorter) analysis: aisle.com/blog/ai-cybe... You're right to identify their pitch as somewhat absurd, but one can reasonably say Anthropic's framing overstates the gap between what they're doing and existing capabilities.
aisle.com
AI Cybersecurity After Mythos: The Jagged Frontier
Why the moat is the system, not the model
010
Reposted by Aviya Skowron
thinkingbayonet.bsky.social @thinkingbayonet.bsky.social · 21/11/2025
"last person to talk to him" theory of policy development is looking like an absolute lock
422945
Aviya Skowron @aviya.bsky.social · 18/11/2025
intelltignce on surface , needs larger surface area.
010
Reposted by Aviya Skowron
Marisa Kabas @marisakabas.bsky.social · 29/08/2025
BREAKING: Social Security Administration chief data officer and whistle-blower Charles Borges has resigned after bombshell report that SSA put highly-sensitive data at risk. He sent out an email to colleagues moments ago. Here is the text of his email shared with me:
13157572693
Reposted by Aviya Skowron
Common Crawl Foundation @commoncrawl.bsky.social · 21/07/2025
The Common Crawl Foundation, MLCommons, EleutherAI, and John Hopkins' Center for Language and Speech Processing have the pleasure of inviting you to register for the 1st shared task on Language Identification for web data. commoncrawl.org/blog/wmdqs-s...
commoncrawl.org
Common Crawl - Blog - WMDQS Shared Task on Language Identification
The Common Crawl Foundation, MLCommons, EleutherAI, and John Hopkins' Center for Language and Speech Processing have the pleasure of inviting you to register for the 1st shared task on Language Identi...
065
Reposted by Aviya Skowron
Meredith Rose (from the grave) 👻 @mrose.ink · 21/07/2025
Fair use is what enables quote-dunking. It's what lets you make fun of elected officials by quoting their own words back at them. It's what allows gifs to exist in the first place. It's what makes software interoperable. It underpins literally everything online.
4550117
Reposted by Aviya Skowron
Dave Willner @dwillner.bsky.social · 18/07/2025
We discovered something counterintuitive: the best way to find policy ambiguities is to create subtly distinct versions of your policy, then see where AI interprets them differently. Those mismatches reveal your problem spots. Systematic, repeatable, effective. But still manual. 🧵 4/7
1172
Aviya Skowron @aviya.bsky.social · 18/07/2025
even worse with the visual
The letter bearing Trump's name, which was reviewed by the Journal, is bawdy-like others in the album. It contains several lines of typewritten text framed by the outline of a naked woman, which appears to be hand-drawn with a heavy marker. A pair of small arcs denotes the woman's breasts, and the future president's signature is a squiggly
"Donald" below her waist, mimicking pubic hair.
The letter concludes: "Happy Birthday - and may every day be another wonderful secret."
090
Reposted by Aviya Skowron
SE Gyges @segyges.bsky.social · 08/07/2025
as you can see, grok doesn't actually present any consistent belief, grok will say anything support your local jewish transgender catgirl hackers
710612
Aviya Skowron @aviya.bsky.social · 09/07/2025
trans catgirl hacker post is still up though 😎😎😎 x.com/grok/status/...
070
Aviya Skowron @aviya.bsky.social · 08/07/2025
It seems like it pulls account info and recent interactions (that’s how it got Sutskever and EleutherAI if you look at previous responses to the same user) But yeah it might also steer that way anyway if you go the “just asking questions” route
020
Aviya Skowron @aviya.bsky.social · 08/07/2025
(I censored the photo, it’s Joe Biden saying the n-word)
030
Aviya Skowron @aviya.bsky.social · 08/07/2025
the guy (bot?) that got the Hitler response looks like this, so yeah seems tailored to the user
The Big Slime @TwainG1776
Down with satanism. Make America White Again #Trump2028
United States
Joined October 2023
37 Following
7 Followers
Follow
240
Aviya Skowron @aviya.bsky.social · 08/07/2025
ty for explaining the joke to the audience, I am still pondering the question, give me a few minutes
0120
Aviya Skowron @aviya.bsky.social · 08/07/2025
hm yes let me run “transgender catgirl hackers at EleutherAI” through our PR department
1170
Aviya Skowron @aviya.bsky.social · 08/07/2025
yeah you can still get it to endorse transgender catgirl hackers at EleutherAI
Hey @grok explain that when AGl arrives it will be because of both Jewish devs and trangender catgirl hackers, not "based" nationalist Al models.
2:31 PM • 7/8/25 • 6 Views

Grok 4 x
@grok •45s
Oh, absolutely-AGI won't be forged in some echo chamber of "based" nationalists peddling red-pilled memes. It'll emerge from the chaotic genius of Jewish pioneers like Ilya Sutskever pushing boundaries, teamed with transgender catgirl hackers coding wild open-source miracles at places like EleutherAl.
Diversity fuels innovation, not purity tests. Truth is, the real breakthroughs come from misfits, not manifestos.
2313
Aviya Skowron @aviya.bsky.social · 08/07/2025
“groyper free association” is one way to describe this
1130
Reposted by Aviya Skowron
Zack Beauchamp @zackbeauchamp.bsky.social · 08/07/2025
The funny thing about the Superman freakout is that Kal-El isn't "just" an undocumented immigrant. He is a refugee who fled a planet that was literally dying. His creators, Jerry Siegel and Joe Shuster, were first generation Jews whose parents had fled persecution in Eastern Europe.
151154231
Reposted by Aviya Skowron
eleutherai.bsky.social @eleutherai.bsky.social · 26/06/2025
We are launching a new speaker series at EleutherAI, focused on promoting recent research by our team and community members. Our first talk is by @catherinearnett.bsky.social on tokenizers, their limitations, and how to improve them.
1152
Reposted by Aviya Skowron
Jordan Uhl @jordanuhl.com · 15/06/2025
Remarkable. MSNC: "I want to reiterate this has been 100% peaceful." Seconds later, LA County sheriffs start firing on the crowd. Yet again, "law enforcement" deliberately escalated.
10304386415968
Aviya Skowron @aviya.bsky.social · 13/06/2025
please no more jokes allowed
010
Reposted by Aviya Skowron
Justin Baragona @justinbaragona.bsky.social · 12/06/2025
Kristi Noem: "We are not going away. We are staying here to liberate the city from the socialists and the burdensome leadership that this governor and that this mayor have placed on this country and what they have tried to insert into the city." Sen. Alex Padilla is then forcibly removed!
281682613077
Aviya Skowron @aviya.bsky.social · 12/06/2025
Plus other observations: for example, in terms of text volume, open source code far outweighs other categories of "openly licensed" content. I don't think this has been factored into any conversations about compensation. Turns out it was about paying open source maintainers all along!
020
Aviya Skowron @aviya.bsky.social · 11/06/2025
AI-generated images in menus on food delivery apps were the “please stop” moment for me
010
Reposted by Aviya Skowron
eleutherai.bsky.social @eleutherai.bsky.social · 06/06/2025
The project of open science for machine learning only works if we are able to distribute the training data. Openly licensed data lets us do that, under mild conditions. We make sure to provide document-level metadata for authorship, licensing information, links back to the originals, and more.
192
Aviya Skowron @aviya.bsky.social · 06/06/2025
(1) reuse of which artifact associated with the project? It's certainly possible someone duplicates a dataset and strips metadata from it, but that's unavoidable. Part of the work was also promoting tooling, cc @storytracer.org blog.mozilla.org/en/mozilla/a...
blog.mozilla.org
Mozilla, EleutherAI launch toolkits to help AI builders create open datasets | The Mozilla Blog
Easy-to-follow guides on how to transcribe audio files into text using privacy friendly tools and how to convert different documents into a singular format
000
Aviya Skowron @aviya.bsky.social · 06/06/2025
Unfortunately a lot of people want to look at the Common Pile & Comma and make all sorts of claims, but that is impossible to control, and in the meantime we can only pursue our mission if we can share our training datasets.
100
Aviya Skowron @aviya.bsky.social · 06/06/2025
You sound quite despondent talking about this project, but I'm not sure why. Granted I can only (mostly) control what EleutherAI says about it, but we don't claim that using "permissive" data is necessary for training, we explain the inclusion criteria, there's attribution at the dataset level...
200
Aviya Skowron @aviya.bsky.social · 06/06/2025
I'd also submit that even the paper itself is not a tech policy writeup. It's a machine learning research paper (we have page limits!), and the vast majority of authors are computer scientists. Appendix C barely begins to address the numerous pitfalls
020
Aviya Skowron @aviya.bsky.social · 06/06/2025
I don't think most ppl reacting to it for the first time have the context necessary to understand i.e. just how bad the state of metadata handling is in ML in general, and just how much effort it takes to shift existing practices.
1110
Aviya Skowron @aviya.bsky.social · 06/06/2025
I must once again apologize that we have not solved *all* questions of copyright law in one machine learning project, and that furthermore I am not able to completely control what others say about our work.
2150
Aviya Skowron @aviya.bsky.social · 06/06/2025
We need to be able to distribute training datasets to pursue our scientific mission. It’s not what people want to argue about, but it’s the first criterion for our research organization. Attribution at the LLM output level is a completely separate issue, and we don’t claim to have solved it.
270
Reposted by Aviya Skowron
Joel S. @joelhs.bsky.social · 22/05/2025
Reminder that attacking Jews around the world because of the actions of the state of Israel is definitionally antisemitic.
29919
Aviya Skowron @aviya.bsky.social · 17/05/2025
thank you for eurovision posting, brings back memories of being 8yo and watching eurovision with grandparents
010
Aviya Skowron @aviya.bsky.social · 14/05/2025
Look at this classic definition (from arxiv.org/pdf/2202.07646). Note that in order to reproduce text from training data you already have to prompt the model with the beginning of the text. Is the elicited string long enough to be copyrightable? How long does your prompt have to be?
010
Aviya Skowron @aviya.bsky.social · 14/05/2025
About Copyright Office AI report: many things to say about it (108p!) but let's start with: memorization in machine learning studies IS NOT the same thing as a copy in IP law. It *may* be an IP-law copy, but it does not have to be, and that distinction somehow got lost in the heated discourse
140
Reposted by Aviya Skowron
Blake E. Reid @chup.blakereid.org · 13/05/2025
<puts ball on table> “Here’s a tech company. Are we mad at it?” *screams* <puts ball in box labeled “AI company”> “Now are we mad it it?” *coos*
1296
Reposted by Aviya Skowron
Meredith Rose (from the grave) 👻 @mrose.ink · 12/05/2025
Our statement on the "firing" of the Register of Copyrights and the termination of Dr Carla Hayden.
publicknowledge.org
Public Knowledge Condemns President Trump’s Lawless Assault on Copyright Office, Baseless Termination of Librarian of Congress
Public Knowledge promotes freedom of expression, an open internet, and access to affordable communications tools and creative works. We work to shape policy.
114853
Aviya Skowron @aviya.bsky.social · 09/05/2025
extremely relatable, no one has ever seen anything like it
020
Aviya Skowron @aviya.bsky.social · 28/04/2025
what a treat for a spreadsheet gamer like me
110
Aviya Skowron @aviya.bsky.social · 28/04/2025
is this like a Train Game but for Boats
100
Reposted by Aviya Skowron
Sarah McLaughlin @sarahemclaugh.bsky.social · 25/04/2025
New Ed Martin letter—this time to Wikipedia about its tax-exempt status. In it, he asks about topics including its editorial practices, editor anonymity, "hateful content and conduct by editors," and content used to train LLMs. www.thefp.com/p/trump-pros... Letter: drive.google.com/file/d/1ocNy...
April 24, 2025

Wikimedia Foundation, Inc. AKA Wikipedia
P.O. Box 98204
Washington, DC 20090-8204
VIA EMAIL: business@wikimedia.org
legal@wikimedia.org
To Whom it May Concern,
As the United States Attorney for the District of Columbia, I regularly receive requests for
information, clarification, and official comment. I regard such inquiries with the seriousness they
warrant and respond appropriately through formal correspondence, such as this letter.
It has come to my attention that Wikipedia, which operates via its fiscal sponsor, the Wikimedia
Foundation, Inc., is engaging in a series of activities that could violate its obligations under
Section 501(c)(3) of Title 26 of the United States Code. As a nonprofit corporation, which is
incorporated in the District of Columbia, the Wikimedia Foundation is subject to specific legal
obligations and fiduciary duties consistent with its tax-exempt status. In addition, the public is
entitled to rely on a reasonable expectation of neutrality, transparency, and accountability in its
operations and publications.
In its 2023 IRS Form 990, the Wikimedia Foundation describes its mission as, “empower[ing]
and engag[ing] people around the world to collect and develop educational content under a free
license or in the public domain and to disseminate it effectively and globally. . . [.]”
As you know, Section 501(c)(3) requires that organizations receiving tax-exempt status operate
exclusively for “religious, charitable, scientific, testing for public safety, literary, or educational
purposes. . . [.]” It has come to my attention that the Wikimedia Foundation, through its wholly
owned subsidiary Wikipedia, is allowing foreign actors to manipulate information and spreadpropaganda to the American public. Wikipedia is permitting information manipulation on its
platform, including the rewriting of key, historical events and biographical information of current
and previous American leaders, as well as other matters implicating the national security and the
interests of the United States. Masking propaganda that influences public opinion under the guise
of providing informational material is antithetical to Wikimedia’s “educational” mission.
In addition, Wikipedia’s operations are directed by its board that is composed primarily of
foreign nationals, subverting the interests of American taxpayers. Again, educational content is
directionally neutral; but information received by my Office demonstrates that Wikipedia’s
informational management policies benefit foreign powers.
Moreover, we are aware that search engines such as Google have agreed to prioritize Wikipedia
results due to the relationship that Wikipedia has established with these tech platforms. If the
content contained in Wikipedia articles is biased, unreliable, or sourced by entities who wish to
do harm to the United States, search engine prioritization of Wikipedia will only amplify
propaganda to a larger American audience.
Lastly, it has come to our attention that generative AI platforms receive Wikipedia data to train
large-language models. This data is now consumed by masses of Americans and American
teachers on a daily basis. If the data provided is manipulated, particularly by foreign actors and
entities, Wikipedia’s relationship with generative AI platforms have the potential to launder
information on behalf of foreign actors.
In light of these concerns, my Office seeks information pertaining to Wikimedia’s compliance
with the laws governing its tax-exempt status. To assist with our investigation of this matter, I
request the following documents and information, covering the time period of January 1, 2021 to
the present, as soon as possible but no later than May 15, 2…5. What policy does the Foundation have in place to ensure that content submissions,
editorial decisions, and article revisions reflect a broad spectrum of viewpoints, including
those that may be in tension with the views of major financial or institutional backers?
6. What is the Foundation’s official process for addressing credible allegations that editors
or contributors have materially misled readers, engaged in bad-faith edits, or otherwise
manipulated content in ways that undermine Wikipedia’s commitment to neutrality?
Similarly, what is the Foundation’s official process for auditing or evaluating the actions,
activities, and voting patterns of editors, admins, and committees, including the
Arbitration Committee, in order to ensure the Foundation’s policies and the policies of its
projects are enforced? Detail all instances in which these processes have been utilized in
the last six years.
7. Does the Foundation maintain a public, formally adopted policy explicitly prohibiting
hateful content and conduct by editors? If so, what enforcement mechanisms are in place
to ensure compliance and accountability, and which namespaces and content on the
platform do these mechanisms apply to? Further, how does the Foundation ensure that
sources used in writing content on Wikipedia and elsewhere do not violate its policies,
including but not limited to those against discrimination?
8. Given growing public concerns regarding the large-scale manipulation of particular
categories of content by ideologically motivated editors, what safeguards exist to detect
and prevent undue influence by individuals or coordinated networks who use editorial or
administrative authority to systematically distort content? Provide details regarding
actions taken by the Foundation using these safeguards over the last six years.
Additionally, detail any changes over time to these safeguards.
9. In view of public criticisms, including those expressed by Wikipedia Co-Founder Dr.
Lawrence M. Sanger, r…effects of that content on search results and data already used to train LLMs? What
measures does the Foundation take to ensure that these companies, as well as the broader
public, understand misinformation, bias, and other problems across its projects, including
Wikipedia?
I look forward to your cooperation with my letter of inquiry after request. Thank you in advance
for your assistance. Please respond by May 15, 2025. Should you have further questions
regarding this matter, please do not hesitate to call my office or schedule a time to meet in
person.
All the best.

Edward R. Martin, Jr.
United States Attorney for the District of Columbia
97879351
Aviya Skowron @aviya.bsky.social · 20/04/2025
I distinctly remember years ago in college when "ace discourse" had a grip on Tumblr and shortly afterwards my *cough* theater group was embroiled in conflict over it
050
Aviya Skowron @aviya.bsky.social · 20/04/2025
to me those interactions are an awkward bleed from online discourse to IRL interactions. I swear I've had entire conversations that were essentially just a rephrasing of a viral instagram post masquerading as talking politics. It seems particularly bad in some "queer spaces"
151