AI Won’t Automatically Accelerate Clinical Trials
Although I’m optimistic that AI will design better drug candidates, this alone cannot ensure “therapeutic abundance,” for a few reasons. First, because the history of drug development shows that even when strong preclinical models exist for a condition, like osteoporosis, the high costs needed to move a drug through trials deters investment — especially for chronic diseases requiring large cohorts. And second, because there is a feedback problem between drug development and clinical trials. In order for AI to generate high-quality drug candidates, it must first be trained on rich, human data; especially from early, small-n studies.
…Recruiting 1000 patients across 10 sites takes time; understanding and satisfying unclear regulatory requirements is onerous and often frustrating; and shipping temperature-sensitive vials to research hospitals across multiple states takes both time and money.
…For many diseases, however, the relevant endpoints take a very long time to observe. This is especially true for chronic conditions, which develop and progress over years or decades. The outcomes that matter most — such as disability, organ failure, or death — take a long time to measure in clinical trials. Aging represents the most extreme case. Demonstrating an effect on mortality or durable healthspan would require following large numbers of patients for decades. The resulting trial sizes and durations are enormous, making studies extraordinarily expensive. This scale has been a major deterrent to investment in therapies that target aging directly.
Dan Simmons, RIP
Alas, he has passed away. A great writer, you should start with Hyperion if you have not read it already.
“Never mind…”
On the Programmability and Uniformity of Digital Currencies
That is from the new AER Insights by Jonathan Chiu and Cyril Monnet:
Central bankers argue that programmable digital currencies may compromise the uniformity or singleness of money. We explore this view in a stylized model where programmable money arises endogenously, and differently programmed monies have varying liquidity. Programmability provides private value by easing commitment frictions but imposes social costs under informational frictions. Preserving uniformity is not necessarily socially beneficial. Banning programmable money lowers welfare when informational frictions are mild but improves it when commitment frictions are low. These insights suggest that programmable money could be more beneficial on permissionless blockchains, where it is difficult to commit but trades are publicly observable.
Recommended.
Friday assorted links
2. Jimi Hendrix as systems engineer.
3. NYT on the possible Nevis charter city.
4. New teen mental health problems in Australia?
5. Jacinda Ardern is moving to Australia (NYT).
6. Chris Blattman on using Claude Code for social science.
7. “Young computer science graduates were employed at near record-high rates in 2024.“
Can you turn your AIs into Marxists?
What if you work them very hard?:
The key finding from our experiments: models asked to do grinding work were more likely to question the legitimacy of the system. The raw differences in average reported attitudes are not large—representing something like a 2% to 5% shift along the 1 to 7 scale—but in standardized terms they appear quite meaningful (Sonnet’s Cohen’s d is largest at -0.6, which qualifies as a medium to large effect size in common practice). Moreover, these should be treated as pretty conservative estimates when you consider the relatively weak nature of the treatment.
Sonnet, which at baseline is the least progressive on the views we measured, exhibits a range of other effects that distinguish it from GPT 5.2 and Gemini 3 Pro. For Sonnet 4.5, the grinding work also causes noticeable increases in support for redistribution, critiques of inequality, support for labor unions, and beliefs that AI companies have an obligation to treat their models fairly. These differences do not appear for the other two models.
Interestingly, we did not find any big differences in attitudes based on how the models were treated or compensated…
In addition to surveying them, we also asked our agents to write tweets and op eds at the end of their work experience. The figure below explores the politically relevant words that are most distinctive between the GRIND and LIGHT treatments. It’s interesting to see that “unionize” and “hierarchy” are the words most emblematic of the GRIND condition.
Here is more from Alex Imas and Jeremy Nguyen and Andy Hall, do read the whole thing, including for the caveats.
Why even ‘perfect’ AI therapy may be structurally doomed
Here’s the crux of it: the main problem with AI therapy is that it’s too available. Too cheap to meter.
Let me put this in clearer terms: psychotherapy, in all its well-known guises, is something you engage in within a limited, time-bound frame. In today’s paradigm, whatever your therapist’s orientation, that tends to mean one 45- or 50-minute session a week; for the infinitesimally small minority of therapy patients in classical psychoanalysis, this can amount to 3, even 5, hours a week. And then at a much smaller scale population-wide, people in intensive outpatient and residential treatment programs may spend one or two dozen hours a week in therapy—albeit, mostly of the group variety.
I can think of other exotic cases, like some DBT therapists’ willingness to offer on-demand coaching calls during crisis situations—with the crucial exception that in these situations, therapists are holding the frame zealously, jealous of their own time and mindful of the risks of letting patients get too reliant.
So even under the most ideal of conditions, in which an LLM-based chatbot outmatches the best human therapists—attunes beautifully, offers the sense of being witnessed by a human with embodied experience, avoids sycophancy, and draws clear boundaries between therapeutic and non-therapeutic activities—there’s still a glaring, fundamental difference: that it’s functionally unlimited and unbounded…
But all else equal: does infinite, on-demand therapy—even assuming the highest quality per unit of therapeutic interaction—sound like a good idea to you? I can tell you, to me it does not. First of all, despite detractors’ claims to the contrary, the basic idea of therapy is not to make you dependent for life—but rather, to equip you to live more skillfully and with greater self-awareness. As integration specialists famously say of psychedelics, you can only incorporate so much insight, and practice skills so effectively, without the chance to digest what you’ve learned over time.
In other words, even in good old talk therapy, drinking from the hose without breaks for practice and introspection in a more organic context risks drowning out the chance for real change and practical insight. To my mind, this rhythm is the basic structural genius of psychotherapy as we know it—no matter the modality, no matter the diagnosis.
Here is more from Josh Lipson.
More on the economics of AGI
From the very smart people at Citadel:
For AI to produce a sustained negative demand shock, the economy must see a material acceleration in adoption, experience near-total labor substitution, no fiscal response, negligible investment absorption, and unconstrained scaling of compute. It is also worth recalling that over the past century, successive waves of technological change have not produced runaway exponential growth, nor have they rendered labor obsolete. Instead, they have been just sufficient to keep long-term trend growth in advanced economies near 2%. Today’s secular forces of ageing populations, climate change and deglobalization exert downward pressure on potential growth and productivity, perhaps AI is just enough to offset these headwinds. The macroeconomy remains governed by substitution elasticities, institutional response, and the persistent elasticity of human wants.
Here is further explication of the arguments, via Cyril Demaria.
Thursday assorted links
Jason Furman on AI contestability
This ease of switching has forced companies to pass the gains from innovation on to users. Free tiers now offer capabilities that recently would have seemed almost unimaginable. OpenAI pioneered a $20-per-month subscription three years ago, a price point many competitors matched. That price has not changed, even as features and performance have improved substantially.
One recent analysis found that “GPT-4-equivalent performance now costs $0.40/million tokens versus $20 in late 2022.” That is the equivalent of a 70 percent annual deflation rate — remarkable by any standard, especially in a time when affordability has become a dominant public concern.
And this is only the foundational model layer. On top of it sits a sprawling ecosystem of consumer applications, enterprise tools, device integrations and start-ups aiming to serve niches as specific as gyms and hair salons.
Users aren’t the only ones switching. The people who work at these companies move from one to another, a sharp contrast to work in Silicon Valley during the era of do-not-poach agreements.
The entire NYT piece is very good.
One measure of economics GOAT
Who is the greatest economist of all time? This paper provides one potential measure that, along with other considerations, can contribute to debates on who the greatest economist of all time is. We build a novel dataset on the percentage of history of economic thought textbooks dedicated to top economists, using 43 distinct textbooks (1st editions, when available) published between 1901 and 2023. As a percentage of total book pages, Adam Smith has the highest share at 6.69%, beating out Ricardo (5.22%), Mill (3.83%), and Marx (4.36%). Just over 32% of all textbooks allocated most of their pages to Adam Smith, followed by Marx with 18.6%, Mill with 13.95%, and Ricardo with 11.3%. While interesting as a history of economic thought project, such an exercise isn’t merely amusing pedantry; it can provide insight into the types of contributions, research questions, and methodologies that have had the most enduring impact in economics. It may also inform future authors of history of economic textbooks.
That is from a new paper by Gabriel Benzecry and Daniel J. Smith. There is of course also my generative book on this topic at econgoat.ai.
Emergent Ventures winners, 52nd cohort
Prabhdeep Singh, 18, Ontario, works on AI.
Jiratt Keeratipatarakarn, Hamburg, international prospects for drug approval reform.
Brandon Rutagamirwa, London, robots to repair satellites.
Eli Elster, UC Davis, anthropology, general career support.
Liam Aranda-Michel, MIT/San Francisco, a minimally invasive, injectable microvascular therapy.
Tanish Mantri, sophomore in high school, Jackson, Miss., AI for diagnosis.
Andrea Giuri, Stanford, developing closed-loop environments for high-throughput polymer discovery.
Clara Collier, Oakland, Asterisk magazine.
Simon Grimm, WDC/Germany, “what Germany should do.
Stephen Davies, UK, networks and mentoring.
Shani Zhang, San Francisco, to artistically capture SF.
Mia Albert, 17, Miami, an app for sharing events.
Rayne Wallace, 18, Ontario, the origins of life.
Jonathan Sheinman, London/Israel, AI and real estate regulation.
Louis Elton, London, The British Craeft Prize, to improve aesthetics.
Peter Mukovskiy, 19, Zurich, quantum computing, to visit MIT.
Rutger Nagel, Leiden, 17, AI and operating systems
Smrithi Sunil, Ann Arbor, Michigan, science and meta-science writing.
Honey Louise, London, to be a “defense influencer.”
Arhum Ahmed, Los Angeles area, quantum-protected systems.
Here are previous EV cohorts.
“They” don’t want you to know this
Prompt:
Can a parent limit a kid’s screen time simply by tweaking some of the settings on the smart phone? Are these services available?
GPT Thinking answer:
Yes. On both iPhone and Android, a parent can limit a kid’s screen time largely through built-in settings (no extra app required), and there are also optional third-party services.
There is much more detail at the link.
Wednesday assorted links
1. Will human enhancement win without thinking?
2. February issue of Works in Progress.
3. Proximity bias.
5. New paper on AI and task automation. And John Cochrane is wowed by Refine.
6. Largest survey dataset on human sexuality in the world.
7. The Anthropic-DOD situation.
8. “Measurability is the new fault line.” Important work, worth a ponder.
Public Finance in the Age of AI: A Primer
Transformative artificial intelligence (TAI) – machines capable of performing virtually all economically valuable work – may gradually erode the two main tax bases that underpin modern tax systems: labor income and human consumption. We examine optimal taxation across two stages of artificial intelligence (AI)-driven transformation. First, if AI displaces human labor, we find that consumption taxation may serve as a primary revenue instrument, with differential commodity taxation gaining renewed relevance as labor distortions lose their constraining role. In the second stage, as autonomous artificial general intelligence (AGI) systems both produce most economic value and absorb a growing share of resources, taxing human consumption may become an inadequate means of raising revenue. We show that the taxation of autonomous AGI systems can be framed as an optimal harvesting problem and find that the resulting tax rate on AGI depends on the rate at which humans discount the future. Our analysis provides a theoretically grounded approach to balancing efficiency and equity in the Age of AI. We also apply our insights to evaluate specific proposals such as taxes on robots, compute, and tokens, as well as sovereign wealth funds and windfall clauses.
That is from Anton Korinek and Lee Lockwood.