Sabitlenmiş Tweet
Evan
1.3K posts
@taobanker Inference is rly slow and expensive without scale. It just isn’t economical to do this
English
Hey so serious question. One of my quants texted me an essay about how he's convinced personal / on-premises computing is about to make a huge resurgence because models are already good enough. My question is -- how far away is that, realistically? How long until retail can get an Opus 4.8 tier model on a $20k machine?
English
The sheer magnitude of this buildout is insane insane insane.
Like, I look at shitcos like $NUAI and I'm like, oh god I hate it.
But it almost doesn't matter, the sheer demand coming from hyperscalers could just force it to work.
Daniel Romero@HyperTechInvest
SemiAnalysis’s Head of Data Center and Energy Infrastructure Research: “Anthropic, we estimate, had 1.5 gigawatts of capacity at the end of 2025. By 2027, they’re going to have 10, so that’s 9 gigawatts in two years. That’s essentially the size of Google today, and you have four companies trying to do that within the same timeframe.”
English
This is correct and a function of hypergamy and Hoeflation. Also in direct contrast to most passport bro destinations.
Westside L.A. Guy@WestsideLAGuy
In the dating app+IG era, very few couples in major U.S. cities (excluding couples who met in college) are ones where the girl is 2+ points more attractive than the guy. Vast majority, guys are more attractive than the girl. Olivia Rodrigo's new BF is an example of this.
English
@Cameron_Invest @ACapitalLP In Q3’25 (when i researched & passed) they had leased 14.6M of the 38M IPv4 addresses at an average monthly price of $0.31, aka $54M annually.
I’m interested in the name but the whole thing hinges on wavelength growth
English
@snapcallcapital @ACapitalLP I'm considering the capital leases to go with the business, more like PP&E. So that leaves ~$1.7B in debt. Plausibly: 38M IPv4 x $35/each = $1.3B. Then data centers maybe $350m + slight excess cash. Yeah it's optimistic but not implausible.
English
I USED MY WHOLE FABLE SESSION LIMIT IN SERVICE OF SCIENCE AND COULDN'T EVEN GET ONE ANSWER
I'M GOING TO MAKE A POST RELEASING MY PROOF WITHOUT CLAUDE'S VERIFICATION NOW
CLAUDE DOES NOT DESERVE A ROLE IN HAVING CREDIT FOR THIS AT ALL
taobanker@taobanker
This is the main reason why there is only one significant math problem that got solved via Fable
English
@Blinklebloop Agree. The entire value of X for an investor rests on ruthlessly curating your feed to only sharps.
Gotta filter the hordes of furus
English
This annoys me tremendously on X.
Gurus will beat their bullish drum over and over again on specific stocks, rallying hordes of followers to pile in.
But nothing is ever said of valuations.
Stock goes down? It’s a buying opportunity- no matter the price it seems.
Valuations do matter, they provide a floor for where actual money starts buying.
If you’re buying at peak valuations, peak multiples, you are just exit liquidity.
English
Fable has verified the core result and Professor Babai has gotten back to me with comments from Bohdan:
This shit is getting real asf. Is it possible I've directed the first significant LLM math proof that isn't a counterexample?
taobanker@taobanker
Professor Babai emailed me back, oh my god I'm scared I really hope I'm not wasting his time 😭 Can someone please have Fable do an adversarial review ASAP?
English
@Cameron_Invest @ACapitalLP Which of their non-core assets cover debt? Datacenters will sell for $500M post-tax in a DREAM scenario against $2.6B debt.
Whole story hinges on wavelength growth, which disappoints each quarter hence the stock price.
English
@ACapitalLP CCOI. Huge fallow asset base. Literally getting paid to take a valuable fiber asset. CEO forced selling. Non-core assets cover debt. Upcoming refinance+revenue inflection. Brilliant CEO...
English
@zeroxkyle Short-sighted. Capex outlay can strengthen moat *and* earnings as long as it has high return on capital. Megacaps will emerge stronger. Buying opp
English
IMO this is one of the best tweets I've seen on this app. Nothing lasts forever - and while the market is far away from even remotely pricing this scenario in, it's here all the same
The scale of the change this presents to the investment case of the Mag7 is pretty insane
Sisyphus@0xSisyphus
Megacaps used to demand high earnings multiples because they were asset light businesses that had moats they didn’t have to spend on This has changed in the last 2 years as these companies need to put hundreds of billions into building data centers to keep a competitive edge
English
While I keep losing money on the stock market, I found myself again learning philosophy🏦 and its "younger daughter," psychology 🧠.
I remember back in Uni my favorite subject was behavioral finance 💰; however, now my interests in studying have slightly changed, yet still keeping the sources peer-reviewed...
Thinking of sharing my thoughts on coping with losses from a psychological and even esoteric standpoint (with healthy skepticism, considering my background) in the format of short videos 📹 - would you be interested in watching❓
P.S.
Why esoteric? A few months ago, I experienced 3 minutes of some out of the blue crazy experience, while being fully sober and familiar with the placebo effect (extremely positive and non-religious or cult-related, though) which basically split my life on before and after. My life vision flipped drastically and I still digesting what was it, but definitely ready to share it.
English
Updated Official Wealth Tiers (2026 H2):
<$5mm: Still poor. Work harder.
$5-$10mm: Spiritual freedom enablement. Choose life over cortisol at all costs or you're simply "the working poor".
$10-$25mm: NBZ (no brownstone zone). You're renting in NYC but your life is good unless you're an insufferable cunt.
$25-$50mm: BEZ (brownstone enablement zone). If owning a brownstone is "making it" to you, you can prioritize ownership but you're really paying for that flex tbh.
$50-$100mm: BCZ (brownstone comfort zone). Careful who you try to keep up with though and don't take any retarded risks.
$100-$500mm: Escape velocity. You're thriving + stacking unless you have Hunter Biden type expenses.
$500-$2bb: SBZ (spare billion zone). Focus shifts towards stacking a bill + having a spare bill for pocket change.
$2-$10bb: FYZ (fuck you zone). Now you're really telling people to fuck off, and they love you for it -- or at least pretend to.
$10+: Retarded zone. Now you gotta think of retarded shit to do just because.
English
Only ones that are worth it and r/r is heavily skewed if I’m right (>3:1 and +100% upside)
Things like META 2022, SE, DLTR, UNH 2025 peak fraud concerns
Not the PYPL’s, cable, LW’s, MTN’s, LULU’s that are trash. Even NFLX right now is dead money no real reason for it to rerate
Dumb Quantum Guy🧃@DumbInvestorGuy
@P_Remarks I feel like this is the only account consistently posting long stonks at 52w lows. Just an observation
English
@HrishbhDalal Totes. It’s still an art, and not great at everything.
3 tweets I’ve found helpful:
x.com/simonlast/stat…
x.com/blader/status/…
Taelin@VictorTaelin
I think I finally figured out how to use AI at scale of course, the fact Fable is good is part of it. but I also changed how I work, and it all comes down to one key realization: you don't need to audit the code, but you NEED to audit the *choices* it made. if you just do that, things will work out, and you'll never lose a codebase to chaos. with Fable at least, the following seems to hold: if I give it a good decision: Fable implements it PERFECTLY if I let it decide instead: Fable may make some bad choices that's how I see Fable: as a perfect execution machine capable of converting good decisions into good codebases, no matter how large. given a concrete plan, it lands its implementation. but when anything is underspecified, it can, and will, make bad choices. that's what you must audit. "while working on this, which choices did you make that you're not confident of? list all." then, you just review that. not the git diff, not 1000's of lines of code. just the choices it made along the way. below is a fresh example. overnight, I asked Fable to fix an issue related to MatMul parallelizing worse than expected. it tracked the culprit with perfection, and landed a solution that DID work. but the solution was not general. it just doubled a buffer, which coincidently fixed the program at hands, but the underlying issue was still present. when it completed the job, it declared success. if I just merged it blindly, the issue would still be dormant. that's the main mistake one can do with AI. instead, I asked it to spell out all decisions it made, spotted the bad one, corrected its course, and now the codebase is clean, correct and the issue is gone for good I really think that if you do that religiously - i.e., NEVER merge without this "which decisions you made?" audit - you can go VERY far without ever reading a single line of code. at least on Bend, this is working incredibly well. despite heavy use of AI to implement an ungodly amount of features I could never dream of, the codebase is still in a superb state, with no signs of degradation fresh example below ↓
English
@snapcallcapital hah sure. maybe skill issue. i am dealing with some arch issues and it overlooks the tiniest of things. can’t share but maybe i will beg it properly next haha
English
fable is not agi, not even close. like i had some basic tasks and fumbles like a clown juggling 8 balls. it gets basic numbers wrong, compares the depth z of non existent prompts to phantom numbers. goes on a wild chase and when i repeatedly ask to check the code fucker goes round and round.
it’s all a hype. yes they can get you started but they are not going to do any sort of research or give you major improvements. if you are getting major improvements, you might have written slop to begin with. which is totally fine. we should find the improvements and learn ourselves. these models don’t have any clue and “oh you are right “ has pissed me off so much that i want to write raw code.
i am getting good at archs but these models are just plain ass. yes they might regurgitate the plane game they memorized or some beautiful front end, but as soon as it gets a bit complicated, flat on the face.
ultimately we are spending lots of time debugging bugs that got introduced debugging bugs with llms. and this might feel productive but lots of time and money is being spent uselessly.
i don’t know about kimi but fable ain’t it. gpt 5.6 is good but not that great either.
maybe i an doing sth wrong. but quite frustrated.
English
Drinking Busch Light in a detached garage while chain-smoking cigarettes will teach you more about deep value investing than four years at Harvard. This is not a joke.
Harvard teaches you what the consensus already knows. The garage teaches you what you can figure out by sitting with a problem long enough for the cheap beer to stop tasting like anything and the cigarette to become the third character in the conversation. One education costs $300,000 and produces a network. The other costs $12 and produces calibration.
Consider what actually happens in the garage. You are alone. There is no professor. There is no case study. There is a folding chair, a printed 10-K on a milk crate, a can of Busch Light going warm in your hand, and the specific quality of silence that only exists in a detached structure attached to a residential property. In this silence you can think in a way that is structurally impossible inside any building where anyone is trying to impress anyone.
Harvard is a building where everyone is trying to impress everyone. That is a description of the incentive structure. If your entire day is spent inside a system that rewards you for looking smart to other smart people, you will develop the skill of looking smart to other smart people, and you will lose the entirely separate skill of being right about something nobody else has bothered to look at. The market pays for the second one and prices in the first one instantly.
The garage does not reward you for looking smart. It gives you four hours of uninterrupted time with a 47-page filing and no reason to perform. In four hours of this you will notice things about a small industrial in central Pennsylvania that a room full of MBAs would miss in a semester, because the MBAs are watching each other and you are watching the cash flow statement.
The cigarette is doing real work. It is a pacing device. You read a page, you take a drag, you read a page. Same technology as prayer beads. It slows the mind down to the speed of the document. A Harvard student reads at the speed of the Socratic method, which is fast and social. A garage guy reads at the speed of a Marlboro Red, which is slow and alone. The slow read is the one that finds the mispricing.
The Busch Light is doing its own work. It is doing the work of not being wine. Wine is a performance. Busch Light is a $0.83 unit of ethanol that lowers your inhibition just enough to let you consider ideas the sober professional part of your brain would dismiss as embarrassing. Some of those ideas are embarrassing. Some of them are the trade.
Harvard will teach you to build a DCF. The garage will teach you the DCF was designed to help you defend a decision you already made to your investment committee, and that the actual work of investing happens before the model, in the moment when you look at a company and decide it is worth understanding.
The Harvard guys are not stupid. Many are brilliant. But they have a structural problem I do not have.
Their compensation depends on other people agreeing with them. Mine depends only on whether the stock goes up. Their information flow depends on other people telling them things. Mine depends on whether the annual report was mailed on time.
This asymmetry is the entire edge. The Harvard guy is being paid to be right in a way other Harvard guys will validate. I am being paid to be right in a way the market will validate over five to ten years. The garage is where the second thing gets made.
I am 35. I did not go to Harvard. I sit in the garage. I read the filings. I drink the Busch Light. I smoke the cigarettes. I buy the stocks. Not all of them work. Enough of them work.
Read the filing. Drink the beer. Smoke the cigarette. Nobody is looking for the mispricing because everybody is inside a building trying to impress somebody else.
I am in the garage. I am impressing nobody.
English