The AI Disclosures Project @ssrc_org
@AIdisclosures
Followers
98
Following
125
Media
6
Statuses
104
Corporate disclosures and standards for AI products and models, led by @timoreilly & @IlanStrauss https://t.co/4Ii3jqPlS0
Brooklyn, New York
Joined July 2024
RT @IlanStrauss: In our most recent @AIdisclosures substack piece: why a chat template standard is important to work towards. If we want AI….
0
1
0
RT @IlanStrauss: Here is a corner of the AI market that you might not of heard of but it matters -- Chat Templates. This is the undocumente….
0
1
0
Arxiv link: *** The #Attribution Crisis in #LLM #Search #Results ***.Estimating #Ecosystem #Exploitation . By Ilan Strauss, Jangho Yang, Tim O’Reilly, Sruly Rosenblat , and Isobel Moure . A collaboration with Prof. Jangho Yang of the @UWaterloo.
arxiv.org
Web-enabled LLMs frequently answer queries without crediting the web pages they consume, creating an "attribution gap" - the difference between relevant URLs read and those actually cited. Drawing...
0
0
0
Want to understand why Cloudflare has intervened to restrict AI crawlers from visiting websites by default?. Read our recent research paper on "The Attribution Crisis" in LLM search systems. To understand how how content is consumed by these AI bots but not cited. Arxiv paper 👇.
2
0
1
In a recent @dataforpolicy fireside chat @IlanStrauss and @srulyrosenblat discuss AI's thinking & behavior. Without a theory of power, mitigations won't amount to much down the line, argues Ilan. A correct "theory of change" is needed, not reliant on simply "more competition".
We @AIdisclosures spoke at a @dataforpolicy fireside chat on AI governance:. I loved being able to test out why we need a "theory of power" . if all this talk of AI governance, risks, and post-deployment monitoring is to build towards systemic change in AI markets.
0
0
1
Big Tech is sealing off user context to entrench AI monopolies. In a piece for @AIFrontiers: @timoreilly , @IlanStrauss & Isobel Moure show how open protocols like Anthropic’s MCP can keep AI markets healthy & open. 👇Read it here: #AI #OpenProtocols.
ai-frontiers.org
Isobel Moure, Jul 30, 2025 — With model performance converging, user data is the new advantage — and Big Tech is sealing it off.
0
0
0
RT @ssrc_org: Next Wednesday at 10am EST, join @dataforpolicy and @AIdisclosures at a Fireside Chat on making AI's thinking more transparen….
0
1
0
RT @IlanStrauss: Today's Asimov Addendum by @AIdisclosures I discuss OpenAI’s telemetry. Metrics, logs, and traces keep AI systems healthy….
0
2
0
RT @ClementDelangue: It’s time for the American AI community to wake up, drop the "open is not safe" bullshit, and return to its roots: ope….
0
80
0
Open-science helps make markets more competitive and more innovative. It allows for best practice to disseminate quickly across the economy. But it also allows for bad practices to be found out. The original "Attention is all your Need" transformer paper was open-science.
In our continued commitment to open-science, we are releasing the Voxtral Technical Report: The report covers details on pre-training, post-training, alignment and evaluations. We also present analysis on selecting the optimal model architecture, which
0
0
0
MCP, as an open protocol created by Anthropic, is having a large impact on the market structure we are seeing emerge in AI applications. How has it transformed your experience in creating a new AI application? We would love to hear from you. Post your story. #MCP #AImarkets.
today we're releasing a new small model (0.5B) for detecting problems with tool usage in agents, trained on 50M tokens from publicly available MCP server tools. it's great at picking up on tool accuracy issues and outperforms larger models
0
0
1
The Reddit discussion on the topic is fun and enlightening as always: .
oreilly.com
A Conversation with ChatGPT and Gemini About Job Losses from AI
0
0
0
"Where Is AI on the Enshittification Curve?" asks our co-director @timoreilly: . "If a company tells you that the principal benefit of AI is 'efficiency,' they are telling you that they don’t see opportunities to create new value or serve new customers." .
oreilly.com
A Conversation with ChatGPT and Gemini About Job Losses from AI
1
0
0
RT @IlanStrauss: Lot of sensible stuff in the White Houses's new AI Action plan. But removing "misinformation" from the most sensible bro….
0
1
0
Design choices matter. Poor citation practices is not a technological constraint. For the full research paper:
0
0
0
Cloudflare has stolen the show for now by allowing websites to block bots -- and even monetize their traffic at a cost. But to what extent are LLMs visiting websites and consuming their content without citing them?. Our latest research shows large differences by model.
1
0
1
In today's Asimov Addendum @IlanStrauss @timoreilly, fixing the #AttributionGap in LLM search. Markets are built on data:.① Devs — ship with full logs.② Publishers — insist on verifiable traces before licensing.③ Enterprises — demand disclosure. More👇.
0
2
3
Read our full audit ⤵ @ssrc_org:. . with @timoreilly @IlanStrauss @janghoYang @isobelmoure and @srulyrosenblat @TechCrunch @BBCTech @bbctechbrief @arstechnica.
ssrc.org
Web-enabled LLMs frequently answer queries without crediting the web pages they consume, creating an “attribution gap” – the difference between relevant URLs read and those actually cited. Drawing on...
0
0
2
*** New WP *** . "The Attribution Crisis in LLM Search Results". AI search assistants are freeloading on the open web. Analyzing 13.9k conversations:. - Gemini provides 0 citations 9️⃣2️⃣% of the time .- Sonar reads ~10 sites but credits 5.- GPT-4o hides much of its trace. 1/2.
1
0
1
RT @ssrc_org: In this paper, @AIdisclosures discusses what protocols could realign incentives towards these creators and address ecosystem….
ssrc.org
Web-enabled LLMs frequently answer queries without crediting the web pages they consume, creating an “attribution gap” – the difference between relevant URLs read and those actually cited. Drawing on...
0
1
0