Pinned Tweet
“We are here on earth to do good for others. What the others are here for, I don't know.”
Afroz Mohiuddin retweeted
In January 2025 my mom’s lung cancer spread to her brain. The drug with the best chance of helping her was approved in China but not here.
A doctor told me, almost as an aside, that if she were Chinese she could buy it at a pharmacy.
So I flew to Shanghai and came home with 26 boxes in my luggage. That was 19 months ago. My mom is still here, still working, and in her words has had an amazing three years.
The drug was approved in China in 2023. The FDA approved it in July 2025. It still is not for sale here. I know of at least 30 Americans who have made the same trip.
I have been scared to tell this story because I worried it could cost my mom access to the medicine keeping her alive. But the problem matters more than my silence. These drugs need to be developed and available here, and that means FDA reform.
Art Allen at KFF Health News told our story this week. It also ran in The Washington Post.
kffhealthnews.org/health-ind…
Afroz Mohiuddin retweeted
There are a lot of things wrong with this world…
but too much intelligence is not one of them.
Afroz Mohiuddin retweeted
if you value intelligence above all other human qualities, you’re gonna have a bad time
Afroz Mohiuddin retweeted
There are a few stories that deserve to be known more widely than that of RLSlow and the ideas that were developed there.
For example — how to do RL at scale with LLMs (variants of which are widely know today) was first developed in that team. And that algorithm was one instance (a special case in fact) of a family of learning algorithms that were studied there.
These ideas were beautiful and deep. And I can’t help but feel that there is a lot left to discover about the nature of learning.
🫡 So much to say here…
Let me take this opportunity to pay tribute to the RLSlow team :) It was a priviledge to lead it through the years, first with @ilyasut then with @merettm and finally on my own - the best team you can ever wish for. It feels like ages ago when we started working on the foundations of reasoning (later berry) with @TrapitBansal, Francis Song, @_smileyball and @sidorszymon.
Many more things happened since then, huge thank you to all teammates through the years: Peter Zhokhov, @unsafePerformIO @vladi_petroff @masonmeyer_ @afrozenator @lukasz_kondr @dmdohan @wang12tao @TheRealRPuri @edwardjhu @eqhylxx @suchirbalaji and Dan Selsam for all the help.
I hope this great technical work continues to be truly and broadly beneficial.
xitter.cf/merettm/status/2096630…
Afroz Mohiuddin retweeted
Let me take this opportunity to pay tribute to the RLSlow team :) It was a priviledge to lead it through the years, first with @ilyasut then with @merettm and finally on my own - the best team you can ever wish for. It feels like ages ago when we started working on the foundations of reasoning (later berry) with @TrapitBansal, Francis Song, @_smileyball and @sidorszymon.
Many more things happened since then, huge thank you to all teammates through the years: Peter Zhokhov, @unsafePerformIO @vladi_petroff @masonmeyer_ @afrozenator @lukasz_kondr @dmdohan @wang12tao @TheRealRPuri @edwardjhu @eqhylxx @suchirbalaji and Dan Selsam for all the help.
I hope this great technical work continues to be truly and broadly beneficial.
xitter.cf/merettm/status/2096630…
I wrote about the state of AI, why I’m concerned about the next few years, and the choices we need to make to keep the future in humanity’s hands.
An Alien Mind: openai.com/index/an-alien-mi…
Learning from the folks on RLSlow, (and more broadly at OpenAI) really made me feel like a "kid in the candy shop", much gratitude @lukaszkaiser and others!
Let me take this opportunity to pay tribute to the RLSlow team :) It was a priviledge to lead it through the years, first with @ilyasut then with @merettm and finally on my own - the best team you can ever wish for. It feels like ages ago when we started working on the foundations of reasoning (later berry) with @TrapitBansal, Francis Song, @_smileyball and @sidorszymon.
Many more things happened since then, huge thank you to all teammates through the years: Peter Zhokhov, @unsafePerformIO @vladi_petroff @masonmeyer_ @afrozenator @lukasz_kondr @dmdohan @wang12tao @TheRealRPuri @edwardjhu @eqhylxx and also Suchir and Dan for all their help.
Let's hope we all can turn this great technical work into something truly beneficial.
xitter.cf/merettm/status/2096630…
There's a new version of this post
Afroz Mohiuddin retweeted
However the exodus of physicists has had a transformative effect on the rest of science: neuroscience, biophysics, and, of course, AI. The technical foundations, intuitions, and creativity brought by physicists has raised the bar across the scientific effort. 2/2
Afroz Mohiuddin retweeted
What if a foundation model could tell us how to modify its architecture to boost inference and reasoning instantaneously—without retraining? What if that tweak incurred near zero latency cost during generation and supported indefinite state tracking? arxiv.org/abs/2608.17981
In investing: You are rewarded not for effort but for the quality of your decisions.
In many areas of life, effort leads to reward. If you study harder, you get better grades. If you practice longer, you can become a better athlete. If you work more hours, you may get a promotion.
The stock market is not one of those places. Instead, you can be rewarded immensely for doing almost nothing.
Herb Wertheim’s early investment in Heico is an incredible example. It is, by almost any measure, one of the greatest Sidecar investments in history.
Wertheim understood something powerful: wealth compounds when your money starts working through other people’s effort. He said:
"You take what you earn… and invest it in other people’s labor."
Most people spend their lives selling their own time, exchanging hours for dollars. While admirable, there is no scale. Wertheim realized the key is to find people who are smarter, more driven, and more obsessed than you, and simply ride on their coattails.
In 1992, he began building a position in Heico when it was a penny stock. An optometrist by trade, running an aerospace company was not in the cards for him. But he didn’t need to. He could just buy a slice of the work being done by the Mendelson family.
Laurans Mendelson, along with his sons Eric and Victor, took over Heico in 1990. They devoted their careers to the company. They lived and breathed aerospace parts, 24/7. They navigated evolving regulatory and competitive landscapes and several recessions. They managed the day-to-day operational grind and made the tough decisions. They built one of the world’s great companies.
While the Mendelsons did the heavy lifting, Wertheim was along for the ride.
And here is the kicker: Today, Wertheim’s stake in Heico is worth $1.2 billion. Remarkably, this is as much as the shares owned by the late Laurans Mendelson himself.
Again, the man who simply bought the stock wound up in the same place as the man who spent thirty-plus years building the company.
This should not be possible and highlights one of the most counterintuitive parts of investing: You are rewarded not for effort but for the quality of your decisions.
The market doesn't know if you did 1,000 channel checks. It doesn't care if you spent 100 hours researching a stock or five minutes. All that matters is if you are right.
It is easy to feel the need to do something to justify our returns. So we have too many opinions, over-trade and constantly tinker with our portfolios. In doing so, we miss the opportunity to build generational wealth by making one or two great decisions and having enough self-awareness to get out of our own way.
An underrated aspect about Codex/CC is how well they abstract away the infrastructure underneath the actual idea you want to bring to life.
I was pleasantly surprised to see Codex just casually bring up a Spark cluster and churn away at things …
A fascinating exchange b/w @patrick_oshag & @ericvishria on youtu.be/0e7aG_MIHlQ
Patrick points out "ironic that ... the returns to being technical are going up even as models supposedly are taking away technical edge."
But good and tasteful research (and engineering) should aim to be vastly broadly more than that, apropos to what @V_Vanhoucke says on this -- xitter.cf/afrozenator/status/208…
Indeed, if it doesn't work in today's model, and if you can afford it, wait for the next one.
Replying to @afrozenator
One consideration is whether your research is itself anchored in time, or timeless by design: are you relying on assumptions that are true today, but are likely to change as technology and humanity progress?
This graph, unfortunately, also mirrors my questions to my colleagues, that are otherwise now answered by Codex.
(1/2)
Replying to @littmath
At the same time certain organs of the mathematical community are atrophying. Below is a graph of MathOverflow questions and answers by month; these numbers have been in slow decline for some time as MathOverflow's function has been cannibalized by Discord etc., but the decline since the beginning of 2025 is likely due in large part to AI. What I find striking here is that there are both fewer questions *and fewer answers*. For example, I was not able to find an increase in answers to older questions in the statistics here, or really any other statistic I could spin as positive.













