Circle IPOs and raises over $1 billion, Apple researchers indicate that reasoning models are still only pattern-matching, and a novel method to purify titanium could turn it into a low-cost material
Great overview thanks Chamath! Lot happening in the world. We already know all the steps that need to be taken. I am starting a new chapter on my favorite book lucky trigger. Love the messages so far.
Always interesting commentary Chamath tho I have to believe we are all dying for a Besties review of the Trump/Elon situation!
Just a comment on the titanium article….its not going to become a base metal until the critical mineral crisis is resolved. Yttrium refining is 80%+ controlled by China and we are 100% import dependent. This type of processing breakthrough will go nowhere until the critical mineral supply chain issues are resolved (perhaps through recycling??)
The solution is not open new mines and refining operations for a taxpayer cost of billions of dollars only for China to tank the price when we are operational and close our mines and operations again. i think it starts with stockpiling and identifying the signals as to when to replenish vs when to release. Titanium will remain on the critical mineral list and not a base metal until Yttrium comes off that same list!
Loved the contrast you’re drawing here between LLMs and reasoning-based systems. I’ve been wrestling with this for a long time—and in my view, it’s not either/or. Neither alone is enough.
It’s tough when you can see what comes next… and the system around you isn’t quite ready for it.
Especially in highly regulated environments, where you're only allowed to build what’s funded—even when the real breakthroughs live outside that roadmap.
It’s the challenge of having a clear vision for the future while navigating a structure built for the present.
And that’s the mindset of real innovation: the courage to question, to imagine beyond constraints.
It’s not just about intelligence. It’s about orchestration, alignment, and evolution at scale.
That next layer hasn’t been built yet—and it won’t come from tweaking what already exists.
One of the more important topics has to be globalization. In school, we're taught how it lowers costs and increases efficiency, but today, we're seeing the downsides of reliance on other countries
Many thanks @Chamath Palihapitiya - the annual letter captures the essence of long-term strategic thinking and adaptability. The reflections on technology paradigms, resilience, and first-principles problem-solving are insightful and compelling…
A fantastic read for founders navigating uncertain times!
Chamath, in your opinion , what is stopping LLMs from learning and internalizing the underlying principles in Physics ? They can understand English , they have been trained to solve basic Math and a fair degree of advanced Math.
Can’t this be taught ? Or said another way, can’t the models be “trained” for Physics ?
LLMs are designed to predict the next word by modeling probabilities and correlations between words, which makes them excellent at pattern matching but bad at deductive reasoning.
That doesn’t mean it is impossible for a probabilistic AI model to discover and then internalize first principles.
The parallel that comes to mind is that humans made observations and looked at patterns in nature over time, which eventually led to the discovery of the laws of physics.
So far, researchers haven’t figured out how to train the models to do this. Despite many orders of magnitude of training data and training time, the models still haven’t figured out first-principled thinking.
That was 44 years ago, per Paul Volcker FED raising federal funds rate to 20% in June 1981 and causing a recession with 10% unemployment in late 1982. A move unlikely to be ever followed, it may be even linked to 1985 S&L crisis and 1987 market crash.
An awesome experience in information. Full mind expansion. Rereading multiple times.
Great overview thanks Chamath! Lot happening in the world. We already know all the steps that need to be taken. I am starting a new chapter on my favorite book lucky trigger. Love the messages so far.
Always interesting commentary Chamath tho I have to believe we are all dying for a Besties review of the Trump/Elon situation!
Just a comment on the titanium article….its not going to become a base metal until the critical mineral crisis is resolved. Yttrium refining is 80%+ controlled by China and we are 100% import dependent. This type of processing breakthrough will go nowhere until the critical mineral supply chain issues are resolved (perhaps through recycling??)
The solution is not open new mines and refining operations for a taxpayer cost of billions of dollars only for China to tank the price when we are operational and close our mines and operations again. i think it starts with stockpiling and identifying the signals as to when to replenish vs when to release. Titanium will remain on the critical mineral list and not a base metal until Yttrium comes off that same list!
Loved the contrast you’re drawing here between LLMs and reasoning-based systems. I’ve been wrestling with this for a long time—and in my view, it’s not either/or. Neither alone is enough.
It’s tough when you can see what comes next… and the system around you isn’t quite ready for it.
Especially in highly regulated environments, where you're only allowed to build what’s funded—even when the real breakthroughs live outside that roadmap.
It’s the challenge of having a clear vision for the future while navigating a structure built for the present.
And that’s the mindset of real innovation: the courage to question, to imagine beyond constraints.
It’s not just about intelligence. It’s about orchestration, alignment, and evolution at scale.
That next layer hasn’t been built yet—and it won’t come from tweaking what already exists.
In my humble opinion.
One of the more important topics has to be globalization. In school, we're taught how it lowers costs and increases efficiency, but today, we're seeing the downsides of reliance on other countries
1. ChatGPT Just Got 'Absolutely Wrecked' at Chess, Losing to a 1970s-Era Atari 2600
https://www.cnet.com/tech/services-and-software/chatgpt-just-got-absolutely-wrecked-at-chess-losing-to-a-1970s-era-atari-2600/
2. A Few Thoughts on NAACP & Trump
https://torrancestephensphd.substack.com/p/national-association-for-the-advancement
Many thanks @Chamath Palihapitiya - the annual letter captures the essence of long-term strategic thinking and adaptability. The reflections on technology paradigms, resilience, and first-principles problem-solving are insightful and compelling…
A fantastic read for founders navigating uncertain times!
Chamath, in your opinion , what is stopping LLMs from learning and internalizing the underlying principles in Physics ? They can understand English , they have been trained to solve basic Math and a fair degree of advanced Math.
Can’t this be taught ? Or said another way, can’t the models be “trained” for Physics ?
Good question.
LLMs are designed to predict the next word by modeling probabilities and correlations between words, which makes them excellent at pattern matching but bad at deductive reasoning.
That doesn’t mean it is impossible for a probabilistic AI model to discover and then internalize first principles.
The parallel that comes to mind is that humans made observations and looked at patterns in nature over time, which eventually led to the discovery of the laws of physics.
So far, researchers haven’t figured out how to train the models to do this. Despite many orders of magnitude of training data and training time, the models still haven’t figured out first-principled thinking.
When looking at juicy treasury returns, keep this in mind:
The highest U.S. Treasury interest rate in the past 40 years occurred in 1981, with the 10-year Treasury yield reaching approximately 15.8%.
That was 44 years ago, per Paul Volcker FED raising federal funds rate to 20% in June 1981 and causing a recession with 10% unemployment in late 1982. A move unlikely to be ever followed, it may be even linked to 1985 S&L crisis and 1987 market crash.
10 year rates since 2000:
Min 0.89
Max 6.03
Median 3.24
Average 3.33