Ilan Strauss (@IlanStrauss) header
Ilan Strauss (@IlanStrauss) profile picture

Co-Director of @AIdisclosures Project Economist affiliated with @IIPP_UCL, @go2uj, and @MSFTResearch

Brooklyn, New York
Joined June 2020
Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)Photo by Ilan Strauss (@IlanStrauss)
How do dominant platforms use their #algorithms to shape markets & extract #rents from suppliers & users? With applications to emerging #AI systems, here are key findings from 3 new papers & 2 years of research with @timoreilly, @MazzucatoM (PI), & @RufusRock2 at @IIPP_UCL 🧵1/
5
21
3
72
30,587
Ilan Strauss retweeted
1/7 Ruchir Sharma: "So far this year, 80 per cent of global equity flows have gone into the US market, double the long-term average. No other country has ever drawn such a large share." ft.com/content/1f1a2a9b-9429…
19
130
9
651
57,422
Interesting interpretation. Here is the study they draw on Hall and Sargent (2011): The debt-gdp ratio fell from 66.2% to 11.3%, a 54.9-point decline. Contributions: - inflation: -34.2 - growth -21.6 - primary deficits -20.8. - nominal returns: +21.7
Did the US really “grow out” of its WWII debt? @AcalinJulien & Laurence Ball find that only a small fraction of America’s post-WWII debt-to-GDP reduction came from growth alone. They argue primary surpluses are needed & caution that the policy mix then is no template for today.
58
Ilan Strauss retweeted
United is blocking AI agents from booking tickets on its system. CEO Scott Kirby says “I don’t want to create another middleman” But the agents can book using an OTA, which is still a middleman and costs United a little more than the agent going direct! wsj.com/tech/ai/united-airli…
55
7
20
276
46,000
Ilan Strauss retweeted
MICROSOFT is no longer the bundler -- they are now the bundled. The biggest threat to Microsoft Office is "Claude Office." Anthropic is dumping free Office products and features to its giant AI customer base. Microsoft has no choice: they must launch a CLAUDE competitor quickly.
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data. Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
430
292
73
4,579
667,841
Happy that my paper w. @LobeckMax has been conditionally accepted at the American Economic Review. We show that citizens have varying beliefs about how inequality affects society, that such beliefs differ from fairness views, and that they causally affect policy preferences. 1/6
22
57
3
314
19,299
Ilan Strauss retweeted
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over. And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams. So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super happy to finally release "The Ultimate MoE Training Handbook", free and open, organized around the questions training teams hit: - how many experts, how big, shared or not? - why is half my step spent in all-to-all, and how do I hide it? - why is my run 25x slower with no error in the logs? - softmax, sigmoid or sqrt(softplus) router, and why keep it in fp32? - aux loss or DeepSeek's loss-free bias? - upcycle a dense checkpoint or train from scratch? - can I serve with fewer experts per token than I trained with? ⚠️ Warning: it's long (~70k words, budget a few days, sorry) but with 55 interactive figures, code you can run and numbers from our own runs. Read it here: huggingface.co/spaces/huggin…
139
623
82
4,560
276,601
Quite a useful graph...
Why Texas really matters. It's set to overtake Virginia as the dominant states for data center. And much of the projects still in early stage (which means, they can be stopped). h/t Derrick Flakoll (BNEF). blinks.bloomberg.com/news/st…
45
Wow. From Raymond James. Trump approval mapped onto average gas prices.
74
My Anthropic IPO valuation got some pushback, which is what it was designed to do. I used that pushback to list out what I believe are the biggest and most fragile assumptions in my story, and examined what can go wrong. bit.ly/4AaMy0P
54
239
18
1,701
184,252
This is pretty crazy... nytimes.com/2026/11/03/us/go… Chromebooks in schools seems to have been an unmitigated disaster for educational outcomes -- a gateway drug to YouTube for teens It's tragic that this data is not available so that a proper econometric estimation cannot be done
JUST IN: Google announces schools that completely ban AI will also lose access to Google Search.
Readers added context they thought people might want to know
Google have just added tighter controls: “controls for Gemini, Gemini Notebook, Gemini in Classroom, and Google Search - access to these products will be entirely off for K-12 student accounts until school administrators explicitly turn them on.” blog.google/products-and-p…
1
50
1/ What do airlines and oil companies tell us about Anthropic's economics? In a new @TechPolicyPress piece, I argue with @timoreilly that Anthropic's IPO filing highlights two major disclosure gaps the SEC can & should address...🧵
1
1
1
2
477
5/ We propose standardized disclosures of input/output tokens, revenue and cost per token, and compute utilization. We also call for greater transparency around major cloud contracts and financial flows between AI labs and their investors, suppliers and distributors.
1
29
Ilan Strauss retweeted
There are many thoughtful critiques of this line of work. This is not one of them. Sad to see @margrev engage in this. Only upside is that the post clearly illustrates the need to have humans in the loop in social science...at least for now :)
9
26
2
252
13,926
Our 2024 paper, led by sruly rosenblat with @timoreilly and me, is out @AIdisclosures We tested GPT-4o, GPT-4o Mini and GPT-3.5 Turbo for evidence of training on paywalled O’Reilly books. GPT-4o: AUROC 0.82 Participatory AI markets need attribution: doi.org/10.1007/s43681-026-0…
1
2
164
This AI hit job adds nothing to this collective work. It treats methodological choices we discuss at length in our own work as “major flaws”. It highlights assumptions discussed in the literature and labels them “errors”. It elevates trivial points as somehow crucial. And it peppers all of this with a torrent of meaningless slop. There is no intelligence in this, only malice. All the experts who have commented on your hit job have found it to be worthless.
5
33
7
464
39,928
Human-pilled? @timoreilly on creating markets for AI in light of the Bitter Lesson: asimovaddendum.substack.com/… ai-disclosures.org/missing-m…
2
115