At @techpolicypress.bsky.social, @ilanstrauss.bsky.social & @timoreilly.bsky.social argue Anthropic must disclose its tokenomics. Like oil & gas, AI labs need standardized token volumes, revenue and cost per token, and compute utilization. www.techpolicy.press/anthropic-ne...

Anthropic Needs to Disclose Its Tokenomics Before Its IPO

The AI Disclosures Project's Ilan Strauss and Tim O’Reilly consider Anthropic's compute bet and the need for enhanced operating metric disclosures.

techpolicy.press

Our 2024 paper, led by Sruly Rosenblat with @timoreilly.bsky.social & me is out We tested three OpenAI LLMs for evidence of training on paywalled O’Reilly books. GPT-4o: AUROC 0.82 Participatory AI markets need attribution @ai-disclosures.bsky.social link.springer.com/article/10.1...

Could AI be more Human-Pilled? In a series of interviews & keynotes co-director @timoreilly.bsky.social explores the meaning of human expertise amidst Richard Sutton's the Bitter Lesson: asimovaddendum.substack.com/p/human-pilled For more on "missing mechanisms" ai-disclosures.org/missing-mech...

Human-Pilled

The Bitter Lesson meets Architectures of Participation

asimovaddendum.substack.com

I was on a really interesting & unique panel: open protocols in AI markets at @presidiosf.bsky.social As an econ I spoke on how open protocols enable different market structures through market-driven specialization. They are @timoreilly.bsky.social market-shaping www.youtube.com/watch?v=5Bb0...

The Importance of Open Protocols | Open Source AI Summit

YouTube video by Presidio Bitcoin

youtube.com

Looking forward to hosting this panel next week Monday at NYU on Open Source and AI Markets, as part of a wonderful workshop on open source AI organized by @creativecommons.bsky.social . It’s open to the public so please register and contribute to the debate: lnkd.in/gnGEm44e

I'm looking forward to moderating this panel of serious thinkers on: what needs to happen for open source to create a more decentralized AI market -- amidst high costs (data & compute requirements)? Come join. You can register here (space is limited): creativecommons.org/event/open-u...

OPEN @ UNGA 81: A Conversation on Collaboration - Creative Commons

On the margins of UNGA 81, leading civil society and digital rights organizations will examine balancing aspirations for digital sovereignty.

creativecommons.org

The AI Disclosures Project@ai-disclosures.bsky.social · 3mo ago

We’re co-hosting OPEN @ UNGA 81 with @creativecommons.bsky.social in NYC on Sept. 21. At 1:30pm, we’re moderating a panel on: Can open source make AI markets less centralized? With Duane O’Brien, Amanda Casari, Madiha Zahrah Choksi & Ben Moskowitz Register: creativecommons.org/event/open-u...

Bigger isn’t always better. LLMs are vastly more general than chess engines like Stockfish -- but at a chessboard the specialized system wins easily. What does chess, LoRA, and small models tell us about the economics of specialized AI? Read our latest: asimovaddendum.substack.com/p/when-small...

When Smaller Models Win

Chess, LoRA, and the case for specialized AI

asimovaddendum.substack.com

Co-director @ilan-strauss.bsky.social on @bbcnewsunofficial.bsky.social commenting on Amazon training its AI models on Twitch creator data The challenge & opportunity, Ilan argues, is to use AI to grow the monetizable ecosystem for all, to benefit creators rather than compete with them Video 1 / 2

I just did a brief @bbcworldnewsrss.newsmast.social.ap.brid.gy interview this morning commenting on the backlash over Amazon using Twitch creator content to train AI. Platforms should use AI to foster win-win outcomes for creators & the broader ecosystem, growing the monetizable pie for all 1/2

AI safety research needs post-deployment visibility We analyzed 1,178 safety & reliability papers drawn from 9,439 generative-AI papers. Just 4% of frontier-lab research addressed high-risk areas of addiction, medical, et Our new open-access paper in @ieeeaccess.bsky.social doi.org/10.1109/ACCE...

Co-director @ilan-strauss.bsky.social asks: "Where does intelligence reside? Where does it live? ...the artificial intelligence found in LLMs is inert & centralized in the models’ weights. This is a very different type of intelligence to that exhibited by markets, which is...." 1/2

Open protocols like MCP pull monolithic AI stacks apart into competitive markets MCP's real unlock is composition across apps, not one server per client New from Adam Jones (@AnthropicAI) & Tadas Antanavicius (@pulsemcp) on how MCP shape's production asimovaddendum.substack.com/p/open-proto...

Where does intelligence reside? I propose that artificial intelligence, as inert & centralized, is reliant on market-based intelligence for federated learning & updating Protocols advance market specialization & can bridge these intelligences: asimovaddendum.substack.com/p/where-does...

Where Does Intelligence Reside?

Why AI needs markets — and how specialized production / open protocols can enable this.

asimovaddendum.substack.com

Really excited to be a part of this cohort. It’s a much needed public-interest research effort on AI & the economy. blogs.microsoft.com/on-the-issue... I don’t believe that AI’s impact on the economy is predetermined...🧵

New cohort of AI Economy Institute Fellows to examine frontier AI firms and the transformation of work

Introducing the 2026 AI Economy Institute Fellows, a global cohort advancing research on AI’s impact on work, firms, and economies.

blogs.microsoft.com

🦜 The Architecture of Participation New in our newsletter: - @ai-disclosures.bsky.social progress toward an architecture of participation for AI markets at Bellagio & globally - @timoreilly.bsky.social: open models, protocols, harnesses & portable memory are now core to AI sovereignty. Read👇

-- We need broadcasting, not "communications" -- My colleague, Sruly Rosenblat, has written a wonderfully incisive piece I think on why A2A protocol hasn't taken off and how we should conceptualize agentic communication. Link below 👇

APIs make markets. That's why paying attention to API policies matter. Sruly Rosenblat breaks down how API policies at the AI Labs are limiting reliability for third-party firms and developers. I wonder if this is helping accelerate the shift to open-source models? Link👇

The AI Disclosures Project@ai-disclosures.bsky.social · 8mo ago

We blame LLMs when they're unreliable. But the model is not always the problem, the API metered access is. A sharp new piece from Sruly Rosenblat @AssimovsAddendum breaks down how API design decisions by a few big companies cap reliability for the entire developer ecosystem. 🧵

We reanalyzed @anthropic.com's Economic Index and found it doesn't show what they say it does: GDP per capita does not predict AI adoption. Moreover, it doesn't imply increasing global inequality, like the @financialtimes.com says. Read @ilan-strauss.bsky.social's full analysis below 👇

The @financialtimes.com headline implied AI adoption patterns portend future divergence in living standards. But Anthropic’s report that they supposedly cover doesn't show this. I reanalyzed their public data @ai-disclosures.bsky.social: the GDP → usage relationship is weak 👇

MCP -- a key open standard in AI application markets -- now lives with the Linux Foundation. MCP underpins a lot of interoperability in the market. But open standards go 'bad' all the time. And it's often not about the standard, per se. Read more in our latest Asimov Addendum 👇

How Open Standards Get Captured

The bundling problem and the thin line between product and protocol.

asimovaddendum.substack.com

With Anthropic reportedly gearing up for an IPO as early as 2026, @timoreilly.bsky.social @ilan-strauss.bsky.social @isobelmoure wrote a “Christmas wishlist” for what its S-1 should disclose based on existing SEC standards. Key is non-price operating metrics which encode business objectives 👇 Read

Our Anthropic IPO Christmas Wishlist

Tell Us What You’re Optimizing For

asimovaddendum.substack.com

With Anthropic reportedly gearing up for an IPO as early as 2026, @ai-disclosures.bsky.social wrote a “Christmas wishlist” for what its S-1 should actually disclose based on existing SEC standards: – burn rate – circular revenue / partnership deals – what algorithms optimize for 👇 Read more

Are LLM's the best that they will ever be? A guest post on Asimov Addendum by @rufusrock.bsky.social "The technology might continue to get better, but that doesn’t mean that the user experience will. Why? Primarily because no one knows how to make LLMs profitable." Read more below👇

@ilan-strauss.bsky.social will be presenting to the Computational Demography Working Group (CDWG) @gohuskies.bsky.social tomorrow (Nov 18) Topic: Can Membership Inference Attacks Detect Paywalled Content in LLM Training Data? Join in! Zoom: washington.zoom.us/meeting/regi... Time: 10-11AM (PST)

In today's @ai-disclosures.bsky.social substack: @timoreilly.bsky.social has an excellent piece on AI as a "jetplane for the mind": AI Is a Tool, Not a Worker -- like all software As AI becomes more autonomous, it must be shaped as a jetplane, not a displacement device, Tim argues Link below

"If Humans Can See It, Bots Can Steal It" In our latest Asimov Addendum post @timoreilly.bsky.social, @isobelmoure, & I look at: - AI bots pillaging the internet - How we got here: The evolving architecture of the internet under HTTP - How we tend to get "there": How business models... 1/2

"If Humans Can See It, Bots Can Steal It" In our latest Asimov Addendum post @timoreilly.bsky.social @isobelmoure, and @ilan-strauss.bsky.social look at: - AI bots pillaging the internet - How we got here: The evolving architecture of the internet under http - How we tend to get there... 1/2

I'm quite pleased that a paper I spent ages on with @timoreilly.bsky.social @mazzucatom.bsky.social was listed in the RUTGERS COMPUTER & TECHNOLOGY LAW JOURNAL's Bluebook 21st ed. of important articles in the field "Amazon’s Algorithmic Rents: The economics of information on Amazon" Links below

Thanks to corporate loopholes, AI giants disclose less than their public peers about their financials and business operations.🧐 But what if we governed AI like a market, not a Messiah? @ilan-strauss.bsky.social + @timoreilly.bsky.social share what could happen ⬇️ techpolicy.press/ai-isnt-a-su...

AI Isn’t a Superintelligence. It's a Market in Need of Disclosure. | TechPolicy.Press

If AI is going to be governed as a market technology, it must be brought into the market’s accountability machinery, write Dr. Ilan Strauss and Tim O'Reilly.

techpolicy.press

AI is not the messiah, its a very naughty market in need of disclosure. Read our most recent Asimov Addendum // @techpolicypress.bsky.social piece for more by Co-Directors @ilan-strauss.bsky.social and @timoreilly.bsky.social We also have a new working paper on SEC Disclosures for AI Below👇