Pinned Tweet
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/
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
72
30,561
Ilan Strauss retweeted
Memory is becoming a core part of agent harnesses/products (e.g. Instinct, @Muse, @cognition Devin, @PrimeIntellect Prime Agent, @NousResearch Hermes, @openclaw, @LangChain Deep Agents, @Letta_AI) -- yet despite every agent claiming to "continually learn", there's no benchmark to actually measure that capability.
A big problem is that memory benchmarks today only measure recall (e.g. simple Q/A about past events). They don't measure if/how agents actually learn from experience to get better at work.
At the same time, benchmarks for coding/knowledge work consist of many isolated, unrelated tasks - so they also don't allow for measuring cross-task learning. The context needed for a task is fully specified up-front (very unrealistic).
In practice, both agents and humans work through sequences of many *related* tasks. Experience from past work provides context and learnings that apply to future work. This is why if you're a manager, you have to provide a lot more context when assigning tasks to a brand new employee or external contractor. Meanwhile, a great senior employee might not need any instructions at all.
With FrontierMemory, we adapt existing benchmark tasks for coding and knowledge work so that instead of all the necessary context being provided up-front and within the task, it has to be learned from past experience. This allows us to compare how well different models and harnesses learn over time.
A big problem is that memory benchmarks today only measure recall (e.g. simple Q/A about past events). They don't measure if/how agents actually learn from experience to get better at work.
At the same time, benchmarks for coding/knowledge work consist of many isolated, unrelated tasks - so they also don't allow for measuring cross-task learning. The context needed for a task is fully specified up-front (very unrealistic).
In practice, both agents and humans work through sequences of many *related* tasks. Experience from past work provides context and learnings that apply to future work. This is why if you're a manager, you have to provide a lot more context when assigning tasks to a brand new employee or external contractor. Meanwhile, a great senior employee might not need any instructions at all.
With FrontierMemory, we adapt existing benchmark tasks for coding and knowledge work so that instead of all the necessary context being provided up-front and within the task, it has to be learned from past experience. This allows us to compare how well different models and harnesses learn over time.
Introducing FrontierMemory: a new way to evaluate agent memory through downstream task performance.
While existing memory benchmarks focus on recall (saturated via agentic search), FrontierMemory focuses on how agents learn from experience.
30
15
214
16,221
Ilan Strauss retweeted
muse is a credit card company.
intelligence will be free...payment processing fees are forever.
intelligence will be free...payment processing fees are forever.
M2020. Top Recaps - Part 1. Stripe is not only the payment engine for Muse, they are actually core to Meta's economics. The 300bps linked card is providing meta with ~100bps of revenue. Card networks are enabling the economics of Agentic. This works beyond Stripe merchants, and beyond Link wallets. The linked card creates alignment with Meta and Stripe. While Stripe may be capable of creating an alternative payment network, they can't create the economics (merchant acceptance of a 300bps card for stripe and non-stripe merchants).
Big retailers are set to block this card (and "wallet") due to price, most of the large merchants require customer auth with merchant. The merchant friendly solution to Muse and agent integration is form @LoginID_io. A fantastic merchant friendly integration to their Auth. Expect a presentation w/ MRC.
20
20
469
47,986
I would agree on the method point: methodological individualism, the classical dichotomy, equilibrium behavior of the system, inability to compare utilities, are all core differences between the neoclassical method and all other schools of economic thought.
The fatal challenge to heterodox economics is that there isn't actually orthodox economics.
Just take a few recent Nobels: behavioral, information failure, institutions, bubbles, financial fragility, labor monopsony (sort of), Schumpeterian creative destruction, etc.
Not to mention RCTs which are emblematic of how much economic research is empirical and only lightly if at all structural.
The real difference is methods. Heterodox economics is an avowedly political project not a social scientific effort to better understand the world. And I have nothing against political projects just shouldn't confuse the two.
0
0
0
75
Ilan Strauss retweeted
I liked Mehrling's online class and books.
There was an exchange between him and Michael Woodford. You can read Mehrling's comments on Woodford's book here: cambridge.org/core/journals/…
Woodford's reply (to Mehrling and other contributors to the Symposium on Interest and Prices): cambridge.org/core/journals/…
There was an exchange between him and Michael Woodford. You can read Mehrling's comments on Woodford's book here: cambridge.org/core/journals/…
Woodford's reply (to Mehrling and other contributors to the Symposium on Interest and Prices): cambridge.org/core/journals/…
Perry Mehrling, RIP.
According to the History of Economics Society, Perry Mehrling died "Monday night from injury sustained in a fall several days earlier." He was Professor of International Political Economy at Boston University and, previously, longtime Professor of Economics was a highly esteemed historian of economics.
Here is his homepage:
sites.bu.edu/perry/
His last book: "Money and Empire: Charles P. Kindleberger and the Dollar System"
amzn.to/47RVhJ2
"Charles Kindleberger ranks as one of the twentieth century's best known and most influential international economists. This book traces the evolution of his thinking in the context of a 'key-currency' approach to the rise of the dollar system, here revealed as the indispensable framework for global economic development since World War II. Unlike most of his colleagues, Kindleberger was deeply interested in history, and his economics brimmed with real people and institutional details. His research at the New York Fed and BIS during the Great Depression, his wartime intelligence work, and his role in administering the Marshall Plan gave him deep insight into how the international financial system really operated. A biography of both the dollar and a man, this book is also the story of the development of ideas about how money works. It throws revealing light on the underlying economic forces and political obstacles shaping our globalized world."
Charles Goodhart: "The discipline of economics is balanced, somewhat unsteadily, between two legs, with the skill of the historian on one side and of the mathematical modeler on the other. In the second half of the 20th century, the tide flowed strongly to elevate the second leg and to diminish the first. In this delightful biography of Charlie Kindleberger, a charming and loveable man, Perry Mehrling shows how Charlie, in his chosen role, as an international monetary economist, fought against that tide, often unsuccessfully. Perry writes so beautifully that one would only need a smattering of economics to enjoy the tale; while if one is interested in the history of economic thought in international monetary economics, then this book is, of course, a must read.'
amzn.to/47RVhJ2
3
21
106
7,990
Interesting on what thinkers are drawn on when you need to understand banking during a crisis.
During the financial crisis, at the Treasury Department, Perry Mehrling, Gary Gorton, and Paul Tucker were the three thinkers whose work was most useful in helping us reach a deeper understanding of the situation. Later, Perry was
0
1
1
86
This Euler turn in macro felt fairly RANK. I much prefer HANK @ben_moll. But macro still feels a bit DANK (acronym TBD).
The Nobel Prize given to Mike Woodford and Bob Hall will re-trigger discussions about the so-called New Keynesian model. Some will hail it as an unambiguous success, building macro from micro foundations. Others will dismiss it as an empirically irrelevant construction, leading to complex models without much relation to reality.
As often is the case, truth is in between. I expressed my views, as a participant in the enterprise, but not an unconditional believer, in an NBER Macro piece last year. Given that the link has expired, I use this opportunity to repost it. shorturl.at/Xq7QG
0
0
1
91
Ilan Strauss retweeted
🌶️ note the rapid access to user logs of family members of a suspected criminal - I'm optimistic that user log access is important in criminal investigations, but we have a long way to go to develop infrastructure for co-maximizing user-privacy at the same time.
while you might w... investigators either get the user logs or they don't... clearly they need to", recent advances in encrypted computation can enable a rather mind-bending level of privacy while also enabling specific queries to be approved in the right context... and with the right checks and balances.
for example, think about how a drug-sniffing dog at the airport protects your privacy. instead of security needing to open every bag to look for drugs/etc... they *only* open bags where they already know there's a problem... and everyone else gets perfect privacy (the dog certainly isn't going to tell anyone about your contents)
encrypted computation can enable this kind of thing for digital systems as well... and it can facilitate checks-and-balances where no single company can unilaterally just "decide" to investigate you and your family...
if you're into technical AI governance... this is the exciting stuff. check out arxiv.org/abs/2012.08347 for more
while you might w... investigators either get the user logs or they don't... clearly they need to", recent advances in encrypted computation can enable a rather mind-bending level of privacy while also enabling specific queries to be approved in the right context... and with the right checks and balances.
for example, think about how a drug-sniffing dog at the airport protects your privacy. instead of security needing to open every bag to look for drugs/etc... they *only* open bags where they already know there's a problem... and everyone else gets perfect privacy (the dog certainly isn't going to tell anyone about your contents)
encrypted computation can enable this kind of thing for digital systems as well... and it can facilitate checks-and-balances where no single company can unilaterally just "decide" to investigate you and your family...
if you're into technical AI governance... this is the exciting stuff. check out arxiv.org/abs/2012.08347 for more
JUST IN: Saudi employee of OpenAI arrested after planting an audio recording device in the company's executive conference room - DOJ
2
6
60
16,975
Ilan Strauss retweeted
@BachFranciske to provide some additional context for the remarks that can be found in this interview.
(2/7) My discomfort with the attention being paid to existential risk---and with the ongoing dominance of the public discussion of such risks by a few so-called thought leaders---is that there's an elephant in the room that is being missed. That elephant is the effect of all of this discussion on young people, and indeed the effect---beginning with social media a decade ago---of information technology on young lives broadly speaking.
(3/7) I recently read an interview by Jill Lepore in which she reports on discussions that she had recently with a group of college students: "They had just three things to say. One, social media completely destroyed their childhoods. Two, AI is destroying their college education and their likely prospects for employment. And three, what the hell are we all doing?" To add a fourth bullet that I've heard in similar discussions: "I'm not going to have children, because the existential risk is just too high."
(4/7) Please contemplate such a statement, those of you who think that it's OK that the public, widely diffused, discussion coming from seemingly authoritative figures is so adamant about existential risk.
(5/7) Let me add that I have little interest in a debate about what "superintelligence" might mean, or whether we're currently seeing the emergence of superintelligence. For what it's worth, I would certainly agree that the solution of extremely challenging math problems (for example) exhibits a very impressive form of intelligence, although I would want to argue that such intelligence is impossible to dissociate from the collective of humans that provided data to the machine learning system (which itself has become a kind of collective).
(6/7) Such philosophical debates are entertaining and even important, but when they then head in the direction of "when such superintelligence emerges, there's a good chance that it will want to kill all humans", I say "science fiction". Again, I don't think that it's bad to consider such thoughts, but I strongly believe that it's bad for society, and especially young people, for such thoughts, unaccompanied by any serious real-world analysis, to become a dominant form of discourse of the era.
(7/7) We are damaging human lives, and the consequences of that damage will no doubt be a major theme of the next several decades.
(2/7) My discomfort with the attention being paid to existential risk---and with the ongoing dominance of the public discussion of such risks by a few so-called thought leaders---is that there's an elephant in the room that is being missed. That elephant is the effect of all of this discussion on young people, and indeed the effect---beginning with social media a decade ago---of information technology on young lives broadly speaking.
(3/7) I recently read an interview by Jill Lepore in which she reports on discussions that she had recently with a group of college students: "They had just three things to say. One, social media completely destroyed their childhoods. Two, AI is destroying their college education and their likely prospects for employment. And three, what the hell are we all doing?" To add a fourth bullet that I've heard in similar discussions: "I'm not going to have children, because the existential risk is just too high."
(4/7) Please contemplate such a statement, those of you who think that it's OK that the public, widely diffused, discussion coming from seemingly authoritative figures is so adamant about existential risk.
(5/7) Let me add that I have little interest in a debate about what "superintelligence" might mean, or whether we're currently seeing the emergence of superintelligence. For what it's worth, I would certainly agree that the solution of extremely challenging math problems (for example) exhibits a very impressive form of intelligence, although I would want to argue that such intelligence is impossible to dissociate from the collective of humans that provided data to the machine learning system (which itself has become a kind of collective).
(6/7) Such philosophical debates are entertaining and even important, but when they then head in the direction of "when such superintelligence emerges, there's a good chance that it will want to kill all humans", I say "science fiction". Again, I don't think that it's bad to consider such thoughts, but I strongly believe that it's bad for society, and especially young people, for such thoughts, unaccompanied by any serious real-world analysis, to become a dominant form of discourse of the era.
(7/7) We are damaging human lives, and the consequences of that damage will no doubt be a major theme of the next several decades.
49
234
1,200
151,791
Ilan Strauss retweeted
anthropic officially letting users use their subs with third-party harnesses via the agents sdk is a massive change of stance and big deal for the agent ecosystem.
why isn't there more chatter around this?
why isn't there more chatter around this?
69
9
1,175
108,146
Ilan Strauss retweeted
i think buzz is criminally underrated. it's basically dots and grok bot, but model and harness agnostic (it can even run omp!). it has a mobile app, lets you share conversations, and even lets you share compute with other people. and it has a really nice slack/discord-like see a lot more hype around it
47
20
269
170,963
It's an interesting paper. The dispersion (variance) warrants greater attention I think.
junior score dispersion roughly doubles in standard-deviation terms, while senior score dispersion declines substantially in the descriptive data.
yet the paper describes...
1/n
junior score dispersion roughly doubles in standard-deviation terms, while senior score dispersion declines substantially in the descriptive data.
yet the paper describes...
1/n
Today we announce new findings from Visiting Fellow David Autor on how AI impacts how professionals build expertise. In a three-month randomized controlled trial with practicing patent attorneys, we test both short-term productivity & longer-term skill building that occur as a resuttps://t.co/3Nw2Uwjaqx
1
0
0
47
...increased senior dispersion despite the opposite direction in Table 2. (The descriptive & regression-adjusted measures seems to differ?)
A more interesting interpretation could be that AI potentially polarizes learning outcomes among juniors, but this requires more tests
2/N
A more interesting interpretation could be that AI potentially polarizes learning outcomes among juniors, but this requires more tests
2/N
0
0
0
9
I was on a really interesting & unique panel on open protocols in AI markets at @PresidioBitcoin
As an economist I spoke on how open protocols enable different market structures through market-driven specialization. They are @timoreilly market-shaping
youtube.com/watch?v=5Bb0r4…
As an economist I spoke on how open protocols enable different market structures through market-driven specialization. They are @timoreilly market-shaping
youtube.com/watch?v=5Bb0r4…
1
4
6
765
Ilan Strauss retweeted
I've wanted this since I was 17.
Since 1991, FRED has made trustworthy and curated economic data free and accessible to the general public. I've been an avid user since my first economics class in high school. A project of @stlouisfed, it's a gem of American public data and ata is now available in FRED!
Since 1991, FRED has made trustworthy and curated economic data free and accessible to the general public. I've been an avid user since my first economics class in high school. A project of @stlouisfed, it's a gem of American public data and ata is now available in FRED!
28
44
450
37,589
Ilan Strauss retweeted
$META Devansh Tandon reveals the tokens-in, engagement-out flywheel behind Meta's recommendation system, where better models lift Reels watch time 30% and the monetization pays for the next training run
"These scaling curves aren't just academic research. They're driving real for some of the biggest consumer businesses in the world."
"Here's a couple of examples I have from Meta's recent earnings reports. Instagram Reels had a strong quarter, 30% year on year watch time."
"And the optimizations we made to improve the quality of recommendations included simplifying our ranking architecture to enable efficient model scaling and longer interaction histories to identify a person's interests."
"We doubled the length of user interaction sequences used for training Instagram and increased the richness of each user interaction."
"So these are direct parallels to the power law scaling curves for LLMs."
"And I want to introduce this idea of a flywheel of tokens in engagement out, which is what's powering all of these recs models scaling."
"You train a model, you then run inference on it, which is the tokens in, that recommendation model results in better content recommendations, it drives consumer engagement, daily active users' time spent, it translates to monetization and ads or subscription, which pays for the next model training run."
"And so every step on the scaling curve is one loop around this flywheel, and a lot of consumer apps are spinning this core flywheel at the heart of their business. So we have a long way to scale these recommender systems."
_______
For the full breakdown of the lecture on Meta's ads recommendation system: firesidealpha.substack.com/p/metas-ad-rec…
"These scaling curves aren't just academic research. They're driving real for some of the biggest consumer businesses in the world."
"Here's a couple of examples I have from Meta's recent earnings reports. Instagram Reels had a strong quarter, 30% year on year watch time."
"And the optimizations we made to improve the quality of recommendations included simplifying our ranking architecture to enable efficient model scaling and longer interaction histories to identify a person's interests."
"We doubled the length of user interaction sequences used for training Instagram and increased the richness of each user interaction."
"So these are direct parallels to the power law scaling curves for LLMs."
"And I want to introduce this idea of a flywheel of tokens in engagement out, which is what's powering all of these recs models scaling."
"You train a model, you then run inference on it, which is the tokens in, that recommendation model results in better content recommendations, it drives consumer engagement, daily active users' time spent, it translates to monetization and ads or subscription, which pays for the next model training run."
"And so every step on the scaling curve is one loop around this flywheel, and a lot of consumer apps are spinning this core flywheel at the heart of their business. So we have a long way to scale these recommender systems."
_______
For the full breakdown of the lecture on Meta's ads recommendation system: firesidealpha.substack.com/p/metas-ad-rec…
Mark Zuckerberg says Meta trains discretion into Muse so it can book a restaurant reservation without disclosing your allergy or pregnancy, something off-the-shelf coding agents skip
"So, I mean, you have to train that into the model. I mean, just like any other capability, it wing, you tell it a lot of things and then you want it to go achieve your goals."
"But let's say you want it to go get you a reservation at a restaurant, and you're trying to find a good restaurant, but maybe you have some allergy. But you may not want to disclose the allergy, or if you're pregnant or something like that."
"But the Muse will know if you tell it. And you're going to want it to be able to do the job but while disclosing as little information as possible."
"So that's a specific skill. And it's something that's kind of basic social skills or common sense that people have."
"But if you're just building a coding agent, most of these companies haven't trained that in. So that's the type of thing that we can do because we're doing the full stack, right?"
"Because we're training the model, we're not just taking someone else's model off the shelf and trying to build a scaffold and an agent around it. We trained the whole model specifically to be good at that."
_______
For the full set of takeaways from Meta's Connect Keynote: firesidealpha.substack.com/p/fireside-am-…
4
27
276
485,441
Why does Opus 5.5 perform so much better within the CLI Claude Code harness than on the web for the same non-coding multi-step task?
Wow.
Wow.
1
0
2
80
Reminds of the Spy vs. Spy comics
🦔The Blue Cross Blue Shield Association says insurers paid out nearly $1 billion more than usual from AI-assisted hospital documentation in 2024-2025. Hospitals use AI to scan records and find secondary diagnoses that human reviewers missed. More diagnoses means higher atment levels haven't gone up.
BCBSA's SVP of data science said AI is "identifying more billable conditions, not sicker patients." Insurers now deploy their own AI to deny those same claims. Patients pay for both through higher premiums.
My Take
This might be the most American healthcare story I've ever read. Hospitals bought AI to find more things to bill for. Insurers bought AI to deny those bills. The technology cost money on each end and the tab went straight to patients through higher premiums. A billion dollars moved and no one got any healthier. The only thing that changed was the size of the invoices.
Two sides of a trillion-dollar industry found a new tool to fight over the same pool of money and the person in the hospital bed pays for all of it. Premiums go up, denials go up, and the AI vendors collect from everyone. If you want to know who AI works for, stop reading the press releases and look at where the money goes. In healthcare it went straight into the billing department.
Hedgie🤗
0
0
0
95
Seems to show me that the EU can design privacy-first product, and it can even build markets and comparative advantage on this basis.
@mariodraghi’s recommendation to foster EU-wide data markets for AI training can be done carefully using
@openminedorg
and TEEs?
@mariodraghi’s recommendation to foster EU-wide data markets for AI training can be done carefully using
@openminedorg
and TEEs?
Spent the weekend with Meta's @Muse, and they got a lot right.
- Every user gets their own VM.
- The agent never sees a password.
- A 2nd agent (Sentinel) has to approve anything that leaves the box.
- Every action is in the audit trail.
That's the most careful consumer agent sandbox I've used.
I don't know if folks realize but a VM for every user, at Meta scale, is kinda huge! It's never been done before where we simply provision a linux microVM for every single consumer.
We recently did this within @Cisco for over 80,000 of our employees. And now within our products as well as we start to scale it out to every single one of our users.
1/n
0
0
1
182