"AI has achieved its strongest results in biology in domains where the
relationship between inputs and outputs is well-defined and heavily
sampled, such as protein structure prediction or de novo antibody design.
These areas benefit from large, high-quality datasets and clear
optimization objectives, which allow models to learn efficiently. But the
availability of these datasets is not accidental: we have good data
because these were problems we already knew how to study and measure well.
As a result, the applications of AI in these domains, while genuinely
useful, mostly accelerate workflows that were already technically
feasible"
15% of men examined for WWII service were disqualified for malnutrition or
its downstream consequences. Related, the average U.S. male height (a proxy
for nutrition)
only plateaued in 1960.
This is a great use-case of LLMs for research. This shapes my mental model
of what LLM-supported college essays could look like in the future.
"Answering 'how early could an airplane have been built?' requires knowing
not just who the various flight pioneers were and when they did their
work, but the state of steam and internal combustion engine technology at
various points in the 19th century and how far they could have plausibly
been pushed. There are just vanishingly few people with this sort of deep
knowledge about even a few technologies, much less 190 of them spread
across two centuries"
"That's not why brand age watches look strange. Brand age watches look
strange because they have no practical function. Their function is to
express brand, and while that is certainly a constraint, it's not the
clean kind of constraint that generates good things. The constraints
imposed by brand ultimately depend on some of the worst features of human
psychology. So when you have a world defined only by brand, it's going to
be a weird, bad world."
This is what I'd like to study in grad school. Two of the authors are from
NYU. Looking forward to this fall.
"This marked a transition in which deep learning theory changed in
character from a largely mathematical study of what is possible to a truly
scientific effort to describe, explain, and ultimately predict the
behavior of complex empirical systems. New scientific endeavors often
start with an empirical tension in which nature presents something
interesting we cannot predict or explain with existing tools, and although
neural networks are artificial computational systems, this same scientific
tension is present here. We should thus approach this task as scientists,
embracing empirics, seeking unifying principles, and identifying recurring
motifs. We should also expect the path forward to look more like the
development of a scientific field than the development of a mathematical
one."
"Concrete goals should include greatly reducing the need for
hyperparameter tuning, giving predictive tools for dataset design, and
providing rigorous foundations for AI safety work."
If the outcome of a prediction market is correlated with a change in the
money supply, market prices stray from probabilities. Separately from this
post, market inefficiencies and asymmetric upsides for long-shot bets are
also cited as reasons to not equate prices with probabilities. Good to know,
and generally still misunderstood I think.
The existence of tax loopholes could paradoxically increase tax revenues by
increasing labor supply.
"In this study, we conducted a series of original real-effort experiments
in an online setting with almost 6,000 participants to test this
hypothesis empirically. Our findings show significant positive labour
supply responses to the opportunity to evade (increased labour supply by
30%). More importantly, the expected tax revenue significantly and
substantially increased by up to 40%"
Online retailers aren't incentivised to allow BYO agents because they would
lose advertising revenue and customer loyalty. As current terms of service
are written it's a grey area, but they could disallow BYO agents already.
This would be bad for competition and consumers (Amazon and Shopify make up
>50% of online retail). It seems wrong to make AI personal shoppers
illegal when human personal shoppers are legal. Retailers have real concerns
like data scraping and bots. Alex proposes a regulatory framework. I agree,
and would side with consumers here. This will change retailers finances a
lot. Claude says Amazon's advertising revenue is 10% of their total revenue
and 25% of their retail revenue.
An opioid-free painkiller with no addiction risk. I'm a little late here,
but seems like this could be a big deal. In 2025 it was approved by the FDA
for acute pain treatment. Still not approved for chronic pain.
