David Benefield retweeted
really astonishing to me, somehow, to see how many math professors implicitly understand the goal of the profession to be providing them with a prestigious paid hobby rather than increasing the amount of mathematical knowledge available to humanity
yes we're all going to lose our jobs and it sucks and i hope we live through it. and also, this is the most exciting time in the history of math
The scale of this is staggering.
But... Why?
Why is OpenAI trying to solve hundreds of math problems to be released all at once? I think we got the point when they found a counterexample to Navier-Stokes -- internal model = great.
I am not sure at this point what this (hostile?) takeover of the mathematical landscape is trying to achieve other than, again, a massive PR stunt.
David Benefield retweeted
there’s so much confusion about what’s happening in AI because if you don’t use coding models every day it’s hard to understand how quickly progress is accelerating but if you do, you likely have AI psychosis to a degree that it’s impossible to explain it in an understandable way
David Benefield retweeted
The 2nd biggest IPO in history is happening under your nose and it has created a unique buying opportunity in memory stocks.
SK Hynix is listing its ADR on the Nasdaq and offering shares to institutional investors. What many have missed is the giga size of $26bn, which would have made it the 2nd largest IPO of all time, after SpaceX. As we saw, markets traded down into the SpaceX IPO. This is mechanical, as funds buying into the offering need to sell other shares to make room.
What is unique about the SK Hynix ADR is that the funding sources are narrow and obvious: Korea-listed SK Hynix as well as Micron, Sandisk, Kioxia, and other similar memory names. The greatest pressure would be on Korea-listed SK Hynix, and my work suggests it could have been up to 20-25% of ADV per day since the offering was announced. Note that foreigners net sold Korean shares for 13 days in a row since the offering was announced, which helps confirm these flows. My sharp friends and I have been searching for the last week to explain the extreme sell-off in memory stocks, as while it felt natural for there to be a cleansing of froth amidst the rolling bull market in AI stocks we’ve seen since last year, it also felt unusually aggressive.
This means it is quite possible $15-20bn of memory stocks have been sold into a narrow exit door, which explains their stunning declines, but also setting up the deck for reversion once these flows cease. You can’t forget that memory is the most important capex ingredient for AI, and everything points towards greater memory demand as LLM can more used, with bigger models and longer context windows. As the offering was over 7x oversubscribed, which is surprising for such a large secondary offering, it is probable that funds oversold and are now underexposed and will have to buy back once allocations are known during the trading day Thursday (likely in the afternoon).
Of course, these flows lined up with exogenous factors that exacerbated the sell-off: frothy and overextended positioning, fears around memory efficiency gains, glimmers of the war restarting, Warsh coming in and being subtly hawkish. All of this came after SpaceX and the end of Q2 rebalance sucked out a lot of flows, which had made the market more fragile. After the stunningly good MU print, I was confused at how poorly it was trading, but now in hindsight it makes sense that is when the institutional VWAPs began. It is logical they would take advantage of the liquidity and demand created post-print to begin exiting in advance of the share offering, and as the negative factors accumulated, they got more aggressive on the offer to try to beat each other out.
The offering is more than 7x oversubscribed: $26bn x 7 = $182bn. This does not mean that $182bn of memory stocks were sold, but there is some relationship between the amount that funds submit for, and the room they need to create. Especially for a secondary offering where it would be more normal to get better allocations versus a hot IPO where getting close to nothing is to be expected. The greatest amount of pressure would have been on SK Hynix (given the ADR will likely trade at a premium), which appears to have then created downwards momentum in Korea, which has been very leveraged, and already seeing other negative factors such as margin increases for retail trading accounts, and rumors of market negative government actions such as unrealized capital gains taxes (which I view as unlikely).
If $15-20bn in the stocks were sold, that is a huge amount of flows to be hitting a few stocks, even if they are large in market cap. As I have explored in a past post about the right tail QQQ leaps, in the current market paradigm, even if market caps are large, free floats are deceptively small as passive flows have crowded them out.
On the efficiency gains, I don’t have any particular insight into the claims although it is very possible they are true. The reality is that efficiency gains around memory are being constantly made. This is not a binary where any efficiency gains mean the end of the trade, rather you can view it as the way that extremely undersupplied buyers of memory are coping with a difficult situation. So even with constant efficiency gains, the situation remains undersupplied, and has a high probability of staying undersupplied. If you look at the memory supply/demand model created by my brilliant friend @AlexCorrino, who has gone deep on these technical nuances, there are good odds the supply deficit will widen much further over the coming years, despite assumptions of significant efficiency gains.
Thus it became an excuse to sell, as memory was already acting very badly due to the pre-ADR mechanical flows, and then when the rumors hit, it created a setup where everyone felt like the efficiency gains were a big deal because the stocks had been acting so bad, and then sellers with different motivations were competing aggressively on the way out.
It was a perfect storm, and it was a perfect storm that you are now poised to benefit from. Generally, these mechanical flows wash out stocks and set them up for reversion, because the baby was thrown out with the bathwater. In this case it is likely there was overselling so those funds will have to buy shares back. And many other holders have sold, and knowing this thesis will give them the confidence to buy it back. It is also probable that some funds that were aware of these flows shorted or sold memory stocks as a trade into that, and will have to buy them back as well.
It’s a good setup, and while I can't say that prices will immediately make new highs now, it feels like a strong and defensible entry point into one of the most important trades of the year, an opportunity that does not come around often. Build medium-to-long term positions and be patient, I suspect it will be rewarding.
Disclaimer:
This is not financial or investment advice, it is not personalized to you, and it does not account for your objectives, risk tolerance, or situation. Nothing here is a recommendation, solicitation, or offer to buy or sell any security, private investment, or related asset, and nothing creates an advisory, fiduciary, or client relationship. The author holds positions in some of the companies discussed here. Positions can change at any time without notice, and there is no obligation to disclose, update, or correct them. The author is not paid by, and has no affiliate or sponsorship relationship with, any company mentioned. Assume a real, built-in conflict of interest, assume the content may be self-serving, and read it that way. Do your own work, check the primary sources, and consult your own licensed financial, legal, and tax professionals before you act on anything. Any decision you make is yours alone and made at your own risk.
David Benefield retweeted
Judging by my tl there is a growing gap in understanding of AI capability.
The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.
But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.
So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.
TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
David Benefield retweeted
service to self whispers “take what you’re owed.” service to others asks “what can i become?” one builds walls around the self; the other opens a door and finds it leads back home.
David Benefield retweeted
Polarity isn’t a war—it’s a dance. Light only sees itself fully when it meets shadow, and shadow only exists to remind light how to move. The trick isn’t to win; it’s to honor both steps of the rhythm.
David Benefield retweeted
People fantasize about hitting the freedom money and then disappearing to live life, but the ones who would actually have disappeared to live life will almost never make it. Why?
Because you don’t wake up one day with 8 figs because you “balanced work and life” perfectly. You get there because you’re a little broken in the head and actually like the game more than the prize. In the same way, bodybuilders don’t quit the gym the day they look shredded.
People with real money all still work in some form. Investing, building, trading, whatever.
When I talk to friends with safe office jobs, they’ll say something like: “If I were pulling what you make, I’d just be on holiday all year.” That’s exactly why they never get there. The second money appears in their head, they’re already planning how to burn it on vacations and luxury items, and end up right back where they started.
It feels a bit controversial to say this, especially from me, Route2FI lol, as you know what the initials stand for, but yeah, I used to think the same way: dreaming of financial independence so I could basically do nothing. Looking back, the problem wasn’t that I hated work. It was that I hadn’t chosen my work. Once you find something you think is fun and you feel good at, you don’t fantasize about quitting. You just want to keep playing.
David Benefield retweeted
Every night human beings go to sleep, lose all sense of their known reality, plunge into an abstract hyper-dimensional realm of infinite experience where time collapses, all moments instantly manifest, and then they wake up and just go about their day.
David Benefield retweeted
Dostoevsky; “People don't want truth, they want comfort dressed as truth.”
Franz Kafka; “Reality is too heavy for most people to carry. So they borrow illusions, soft dreams, sweet lies, and call it happiness.”
David Benefield retweeted
When unexpressed love has a specific soul's name engraved upon it, it is reserved in totality for them alone. The unspoken truth is that it cannot be given to a substitute, and so it remains in the place where ephemeral whispers swim. Thus, the greatest blessing in life is when the one who carries such love is able to give it in full to the one it was meant for.
David Benefield retweeted
In 1945, Solzhenitsyn was a decorated Soviet officer who made a small, private joke about Stalin in a letter.
The state opened it, read it, and treated it as a crime. Within weeks he was arrested and stripped of rank. He was fed into the camps, and sentenced to eight years in the Gulag.
The camps were designed to teach one lesson: say nothing, remember nothing, become nothing. He shoveled frozen concrete until his hands split and bled.
Years later, Solzhenitsyn would write, “Bless you, prison, for having been in my life.” It sounds insane until you understand what he meant. Prison showed him the truth of the regime in its purest form.
After his release, the punishment did not end. He lived under constant surveillance, moving from place to place, knowing that writing a single page could mean death. So he did not write. He memorized. Whole chapters of The Gulag Archipelago lived only in his head. Friends hid scraps of text. Wives memorized passages. For years the book existed only in human memory, as fragile and dangerous as a secret prayer.
When it was finally published, it did not argue that Soviet communism had gone too far. It showed that this was exactly where it led. Solzhenitsyn had learned that systems built on lies survive only if people agree to repeat them, and that the simplest refusal… to stop saying what you know is false… is the first and most dangerous act of resistance.
David Benefield retweeted
forgiveness isn’t letting someone off the hook—it’s realizing there was never a hook, just your own hand clenching a memory until it calcified into identity. to forgive isn’t forgetting. it’s composting. it’s alchemy. it’s choosing to stop drinking the poison to spite the ghost.
David Benefield retweeted
The biggest story of last year, and probably the biggest story of the coming year, is the widespread discovery of the fact that we have never been more accountable to government, which can see into our lives with unprecedented clarity and monitor our utterances to the word, while it has never been less accountable to us.
On the verge of Artificial General Intelligence, so we're told, we still can't count votes, police billion dollar frauds, penetrate the chambers of elite decisionmaking, or figure out where much of our money is going and what's being done with it. This is the kind of crisis that undermines the basic legitimacy of nations and regimes. How we deal with it, or don't, will determine our fate.
David Benefield retweeted
time doesn’t pass, we do. shredding through the eternal now like kids with crayons, coloring inside illusions because infinity without structure makes us panic. the calendar isn’t real—it’s just a security blanket we wrapped around the void.
David Benefield retweeted
Replying to @theralkia
All the tension of every withheld truth settles in the body.
David Benefield retweeted
you didn’t come here to feel good, you came here to learn how to transmute discomfort into presence. earth is a mystery school—not a spa. bless the triggers, they're your real instructors. enlightenment isn’t a glow-up, it’s a slow burn.
David Benefield retweeted
Time isn't a line you follow but a canvas you're continuously painting on. The masterpiece isn't some far-off goal—it's the present brushstroke, where eternity peeks into now. The magic? It's in realizing you're the artist and the art, creating time by living it fully.
David Benefield retweeted
I took delivery of a beautiful new shiny HW4 Tesla Model X today, so I immediately took it out for an FSD test drive, a bit like I used to do almost daily for 5 years. Basically... I'm amazed - it drives really, really well, smooth, confident, noticeably better than what I'm used to on HW3 (my previous car) and eons ahead of the version I remember driving up highway 280 on my first day at Tesla ~9 years ago, where I had to intervene every time the road mildly curved or sloped. (note this is v13, my car hasn't been offered the latest v14 yet)
On the highway, I felt like a passenger in some super high tech Maglev train pod - the car is locked in the center of the lane while I'm looking out from Model X's higher vantage point and its panoramic front window, listening to the (incredible) sound system, or chatting with Grok. On city streets, the car casually handled a number of tricky scenarios that I remember losing sleep over just a few years ago. It negotiated incoming cars in tight lanes, it gracefully went around construction and temporarily in-lane stationary cars, it correctly timed tricky left turns with incoming traffic from both sides, it gracefully gave way to the car that went out of order in the 4-way stop sign, it found a way to squeeze into a bumper to bumper traffic to make its turn, it overtook the bus that was loading passengers but still stopped for the stop sign that was blocked by the bus, and at the end of the route it circled around a parking lot, found a spot and... parked. Basically a flawless drive.
For context, I'm used to going out for a brief test drive around the neighborhood to return with 20 clips of things that could be improved. It's new for me to do just that and exactly like I used to, but come back with nothing. Perfect drive, no notes. I expect there's still more work for the team in the long march of 9s, but it's just so cool to see that we're beyond finding issues on any individual ~1 hour drive around the neighborhood, you actually have to go to the fleet and mine them. Back then, I processed the incredible promise of vehicle autonomy at scale (in the fully scaleable, vision only, end-to-end Tesla way) only intellectually, but now it is possible to feel it intuitively too if you just go out for a drive. Wait, of course surround video stream at 60Hz processed by a fully dedicated "driving brain" neural net will work, and it will be so much better and safer than a human driver. Did anyone else think otherwise?
I also watched @aelluswamy 's new ICCV25 talk last week (nitter.cf/aelluswamy/status/1981…) that hints at some of the recent under the hood technical components driving this progress. Sensor streams (videos, maps, kinematics, audio, ...) over long contexts (e.g. ~30 seconds) go into a big neural net, steering/acceleration comes out, optionally with visualization auxiliary data. This is the dream of the complete Software 1.0 -> Software 2.0 re-write that scales fully with data streaming from millions of cars in the fleet and the compute capacity of your chip, not some engineer's clever new DoubleParkedCarHandler C++ abstraction with undefined test-time characteristics of memory and runtime. There's a lot more hints in the video on where things are going with the emerging "robotics+AI at scale stack". World reconstructors, world simulators "dreaming" dynamics, RL, all of these components general, foundational, neural net based, how the car is really just one kind of robot... are people getting this yet?
Huge congrats to the team - you're building magic objects of the future, you rock! And I love my car <3.


























