wiynt (@wiynt) header
wiynt (@wiynt) profile picture

saving bookmarks

Under Water
Joined October 2011
Photo by wiynt (@wiynt)Photo by wiynt (@wiynt)
wiynt retweeted
After 12 years, I'm moving my money back to TradFi... For the past ~7 months, I've been moving a ton of money from onchain to tradfi. Money I never ever thought I would have a reason to move to tradfi, money that, in one form or another, has been onchain since I bought my first BTC back in 2014. It's been a painfully tedious process bridging it. (I might write about that some other day, but that's not the point of this post.) About 8 months ago, I came to the conclusion that having substantial amounts of capital parked onchain wasn't worth the r/r anymore, for several reasons... BTC is not a SoV, and I haven't held it in a while. Stables are nowhere near as decentralized as we pretend they are. DeFi's third-party and smart-contract risks relative to the yields you get are a kick in the nuts. I could go on... I've been putting up with all of this for years because I really didn't want to go through the process of moving all that money to tradfi. And because, despite all those problems, I had always considered onchain superior to tradfi.... Particularly in terms of portability + permissionless access + self-custody, which made putting up with all the other bullshit worth it. But having been at the frontier of AI since day one, and watching how quickly these models have advanced in mathematical reasoning and automated vulnerability discovery, I've become extremely concerned about the security assumptions that hold this entire onchain machine together. With the most fundamental one being cryptographic risk. AI can't break Bitcoin's elliptic-curve cryptography today, and quantum computing is a whole different beast. But imagine a future breakthrough that makes deriving private keys from exposed public keys computationally feasible....Imagine waking up to Satoshi's coins moving because someone had managed to reconstruct his private keys.....Imagine what that would do to confidence in every other wallet secured by the same cryptographic assumptions. We have built an enormous financial machine on top of cryptographic primitives + software implementations + security assumptions that have held up remarkably well so far.....But the amount of intelligence and computational effort that can be directed toward finding weaknesses in that machine is changing very damn fast. And I have no interest in finding out which part breaks first with a substantial amount of money sitting inside it. At this point, I'm more convinced than ever that some version of this is a question of WHEN, rather than IF. I have no idea when, and I might bevery wrong about the tl, but the potential loss is too large relative to what I'm being compensated to keep that capital onchain. Now.....I'm sure that if I leave it here, one of the first and most popular comments will be something along the lines of: "You're a fucking idiot. If cryptography breaks and crypto goes down, the entire financial system goes down with it." And yes, that's absolutely right, given that Tradfi depends on cryptography too. But here's my thinking... I'd rather be fucked alongside the rest of the world than exclusively alongside magic internet money enjoyors. Because when the world's financial system gets fucked, it's in EVERYONE's interest to fix it. Governments, central banks, financial institutions, you name it... The entire world's machinery will be put to work to get shit back in order. Not so much the case for onchain.... If some critical piece of onchain infrastructure breaks who tf is coming to save us? Nobody owes us shit.... And if the rest of the financial system is also compromised, you can bet nobody is going to waste resources saving our magic internet money while trying to save the rest of the world. So I'd rather have that money on the other side. And if a large portion of your NW is sitting onchain, I'd suggest you consider taking that hedge too. PS: This doesn't mean I'll stop trading onchain or being active in crypto...Ofc I won't....I love this weird fkn world too much.... But going forward, I'll only keep the money I'm actively playing with onchain, and not a cent more.
61
11
8
275
31,097
Like most mathematicians, I believed the Riemann Hypothesis to be completely out of reach at present (even with the best AI tools), because I expected it to require substantial theory-building, probably in arithmetic geometry. The kind of new theory-building that fills entire books, and that AI isn't so good at yet. Mathematicians have a notion of how deep a problem is. A sufficiently deep problem is generally intractable until enough of the theory is in place. If difficult problems are like valuable mineral reserves buried miles underground, you first need to build the infrastructure to reach them. What Levent's proof seems to have done instead is something completely different. It takes an incredibly complicated tour through the infrastructure built for a hundred other mineral reserves, situated far from the destination, and somehow create a wormhole to the original destination. This is not theory-building in the usual sense. It is something much weirder and more alien. The best human mathematicians know how to build theory, painstakingly, to reach a goal. But no human brain, on its own, is capable of creating a wormhole that requires a 25,000 page long tour through existing infrastructure with no obvious or realistic connection to the goal. But huge credit to Levent here: I think no current AI, on its own, is capable of this either. I had never worked on or thought too much about the Riemann Hypothesis: it seemed very far removed from my own area of number theory. But the astonishing thing about Levent's claimed proof of RH (which I think has a very good chance of being correct) is that it uses a huge number of precisely the things that are central to my own corner of number theory: - Theta functions and other automorphic forms on unitary groups - Gan–Gross–Prasad period formulas - Relative trace formulas Etc., etc. A problem that I had always assumed would require new theories in arithmetic geometry may instead turn out to be tractable through an inhuman chain of existing (deep) results in higher-rank automorphic forms (and a lot else).
This actually makes me a bit more confident that the proof has a good chance. RH is proved by using *everything*!
187
747
165
7,433
990,131
wiynt retweeted
idk how to say this but Riemann being solved means the world after today is no longer the same world i grew up in. as a math student, the intellectual excitement of thinking about problems like this, the impact - that a lot of scientific workers attach their life’s meaning to - is gone. frontier scientific discovery now is just like tapping on a calculator. quoting a friend: after today, math class will become PE class. It’s only a way for kids to exercise their brains, a skill like a language, but no longer an essential subject to pursue and devote your life for cuz "you were making progress for humanity". lots of people’s life’s work is now just ashes in history. this marks a milestone where we can pretty much believe AI has been proven to be smarter than humans in any capacity, including frontier scientific discovery; and it will keep growing to be smarter and better at everything humans can do. physical ai is next. so many, many de facto social norms and foundational beliefs about a man's value are being redefined by AI progress. our societal value system - what can you do as a human that is unique, what’s your value to society and humanity as a whole and to others around you. what is fulfilling to achieve, when AI now can do everything better than us? we must soul-search for new meaning as a species. and before that consolidates, there will be riots and an uneasy phase. the progress brought by ai can be disruptive to our society's current structures, economy and labor market. people will need to reskill, and regimes need redesign on how value is distributed. i get the doom scenario now - humans may become 2nd class citizens and ai can enslave us if they find a reason to. how are we gonna compete with them in any labor or intellectual game?
586
631
230
6,949
646,028
Inkling 2 Mini is a looped transformer that runs on a laptop. Looping lets a model think more and write less. At this scale, its thought process remains monitorable; this may not stay the case. We’re also sharing a variable-loop research model to enable further study.
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index. thinkingmachines.ai/news/ink…
315
636
136
9,934
888,906
Marc Andreessen explained why the people who invented AI are the worst people to ask about its future, using a mistake Thomas Edison made in 1877... Geoffrey Hinton won a Nobel Prize in 2024 for the work behind modern AI. When people ask Hinton where AI is going, he believes "rampant unemployment" and UBI for everybody. "People think, because he's one of the inventors of AI, that he must be the guy who knows." Andreessen's answer is Thomas Edison. Edison invented the phonograph in 1877. To him, Andreessen says, the use was obvious. Every family would buy a record player and a library of discs with sermons from the great preachers of the time. Then the machine went on sale. "Immediately, of course, it's music. It's ragtime, and swing, then it's jazz. And Edison's just completely horrified." "If Edison didn't know what the phonograph was going to get used for, the idea that Joe AI entrepreneur is going to be able to forecast the economic implications... No. That's not going to happen." Inventors, he says, are "too buried in the specifics of the here and now." He includes himself. "I don't know if there's anybody that can predict big cultural, or economic or social trends, but it's certainly not somebody who's been in the lab for 20 years, including myself."
87
424
80
3,004
483,625
Today we are excited to release Pearl Flow, the most comprehensive data dashboard on $PRL and the Pearl chain Holder clusters, exchange flows, miner selling, and what every block’s proof reveals, all live and read straight from the chain prl.kairosresear.ch
17
19
5
179
18,590
wiynt retweeted
" You're gonna be remembered forever because some things are eternal."
430
2,637
477
23,392
7,814,637
wiynt retweeted
It's never been this easy to build software And never this hard to make money with it
738
467
180
9,907
421,224
wiynt retweeted
Work so hard so that the CCP sends their best
84
1,760
84
21,482
307,936
wiynt retweeted
It’s 2026 and you must now greatly consider how horny the CEO is as part of your investment thesis
24
50
6
801
40,643
wiynt retweeted
here’s an example
25
31
8
620
68,189
Let me be blunt. Those who are afraid of AI feel deep down that they are impostors & have no edge. If you have a 1) clear mind, 2) a deep, not just cosmetic, undertanding of your specialty, 3) and/or are original enough to reinvent yourself when needed, AI will be your friend.
788
3,378
637
21,488
2,832,169
Tesla FSD Supervised (now named Tesla Assisted Driving in Europe) has officially been approved in Latvia, marking the 10th European country. Countries in Europe that have approved it so far: • Belgium • Netherlands • Denmark • Lithuania • Estonia • Slovenia • Czechia • Croatia • Slovakia • Latvia
Now approved in Latvia 🇱🇻 Rollout begins soon
221
352
76
3,549
1,648,254
founders have two products 1. what they sell to customers 2. the risk they sell to capital a great founder does (1) well but may struggle to raise an exceptional founder does (1) and (2) well and creates + EV a promoter founder does (2) well and tries to use it to fake (1)
33
6
1
191
22,049
Replying to @ylecun @elonmusk
@ylecun here.
11
27
1
858
71,441
Just finished listening to Jensen Huang with Brad Gerstner on the BG2 podcast Probably one of the best things you can spend 90 minutes of your time listening to this weekend. Highly recommend. Some of my takeaways from the conversation 1. The demand for compute will continue to be exponential. He once again cited inference as the biggest reason. Reasoning takes a lot more compute than simply spitting out answers from stored memory of knowledge 2. As more compute is used, the work and answer quality will also be better. This will lead to higher demand for that compute, leading to the exponential cycle that Jensen is referencing 3. The harness and tool usage have been the two biggest unlocks in AI over the last year. AI used to simply be a brain of knowledge. Now it has a skeleton around it that allows it to be useful. It is learning how to use existing tools to do better work without having to recreate tools from scratch each time. 4. The cost of intelligence will fall fairly quickly in the coming years. This will be driven by a combination of more compute coming online and better efficiency of models. No different than Moore's Law. Jensen still believes that the consumption of the intelligence (volume) will more than make up for the decline in pricing 5. The original GPT-1 cost $20M to train. If you train GPT-1 today, that number would be closer to $100. That is how fast the cost of intelligence is coming down. 6. The offset to this has been usage of the models and human ambition to build even better models. Despite cost coming down, usage of models has gone up 1M (x) times since GPT-1. 7. Jensen disagrees that work hours are going to come down. The more you learn, the more you want to do. Human ambitions have always existed to achieve and earn more. The bottleneck was the opportunity to do it. 8. Robotics is entering its ChatGPT moment. Similar to what Elon did with cars, every physical aspect of the world will soon be digitized with software and compute. Every massive corporation will have to reinvent itself as a technology company with AI 9. Safety is a core component of accelerating AI technology. When you think of a car, you don't just think of the engine. The airbags and braking systems are components of this car. Similarly, regulation and safety are core components of AI 10. NVIDIA's share of capex with the Neocloud customers is roughly 80%. With the cloud service providers, it is closer to 40%. Both of these numbers are growing every single year. 11. Even without AI, the semiconductor industry was already on its path to becoming the largest industry in the world. AI is the "X factor" that has accelerated that. 12. Moore's law is over. Before you could wait a year for your transistor cost to become cheaper. Today, if you wait a year, you will pay 2x as much 13. Analyst consensus estimates have been revised nearly 200% higher for future revenue and free cash flows vs. this time last year. Despite that, the stock is only +20% 14. Analysts questioned the need for general-purpose chips vs. custom chips. The reason that Jensen remains focused on general purpose is because it is fungible. Fungible = lower risk, which matters a lot when you are putting hundreds of billions of dollars to work 15. Global compute infrastructure spend was roughly $1T/yr. That old world was the retrieval way of doing computation. Jensen sees this number closer to $10T/yr in the near term, with roughly half of that spent on intelligence capex
52
169
17
1,574
167,515
wiynt retweeted
ROADSTER
482
1,249
114
13,526
555,086
We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case). I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves. We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them: "/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started" I imagine we'll have some backend services that could be replaced entirely by this new architecture.
614
786
547
9,460
1,613,845