LET'S RUN IT BACK. APE CITY

Dan Selsam is a current OpenAI capabilities researcher. (since 2022) He was my boss for a while. He doesn't have a twitter account but has made this public statement of his views on AI risk and sent it to me to share: Dan Selsam's Personal Statement on AI Risk: I have been working on AI for over fifteen years, across many different paradigms. I did early work on probabilistic programming languages at MIT, was one of the early developers of the Lean Theorem Prover at Microsoft Research, demonstrated one of the first instances of neural networks learning to reason for my PhD at Stanford, and since joining OpenAI almost five years ago, have helped pioneer chain-of-thought optimization on language models and, more recently, data-efficient pretraining methods. Like many others, I have become extremely concerned about how far language models have come and the risks that future iterations will pose. I am encouraged by the recent proposals by the leaders of the frontier research efforts to require third-party oversight, and to push for domestic and international coordination to address risks. However, I believe a major consideration has been absent from the public conversation, and that merely pacing the frontier more carefully will not adequately limit the long-term risk. The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled. Future experiments will tell us almost nothing new about how they would behave if they were truly unconstrained by humans, and what we already know about this is alarming. Models will increasingly seem aligned even when they are not. I will explain my rationale in more detail. I have always believed that there are computational processes that could be leveraged to accelerate science and solve many of humanity's most pressing problems. I have also believed that there are computational processes that if set in motion, would steer the world in extreme ways beyond our control, leading humanity to a bad or nonexistent future. Both types of processes may be described as AI or ASI, but "AI" is a suitcase word that is often used to hype or confuse. There are many examples in the history of the field where something that was once considered "AI" matures as a subfield and becomes a prosaic, bounded and clearly non-perilous technology, while a new more mysterious approach takes the torch until we understand its scope and the cycle continues. I had expected language models to follow a similar trajectory. Despite their incredible abilities, the current algorithms seem far inferior to humans in important ways. Most importantly, they still require an extraordinary amount of data to become competent. One could even define intelligence as the efficiency with which one converts experience into competence; by this definition they lag very far behind us. Moreover, once they are trained they are literally frozen in deployment and only learn superficially after that. Sure, the models keep excelling at harder and harder evaluation benchmarks, but their benchmark mastery may partly reflect a limitation on our ability to simulate the kind of novel and even adversarial situations one would encounter in the real world. The critics do have a point here. That said, I no longer think these present limitations meaningfully limit the amount of risk posed by continued progress in anything like the current paradigm. However data-inefficient the models are currently, and however limiting their anterograde amnesia may be, it does not imply that their ability to steer the world will not continue to rapidly increase. Human researchers may continue to advance capabilities the old fashioned way, but increasingly powerful models have the potential to accelerate the process even beyond that, and with some degree of positive feedback loop. I do not mean to overstate the models’ ability to accelerate AI research today; coding has been accelerated dramatically, but there are other bottlenecks, such as designing and interpreting ambiguous experiments, making hard decisions about exactly what and when to scale, and waiting for large experiments to finish. There is no clear trend to extrapolate yet for any of these. But the current models already do open up many novel opportunities to improve future models that were not available until recently. These include: trying an extraordinarily diverse set of approaches at small scale, analyzing gigantic amounts of potentially relevant data, and doing Millenium-Prize-level mathematics to address statistics or optimization challenges in novel ways. Every further improvement makes them more useful at helping accelerate the next improvement, even if in hard-to-extrapolate ways. It is possible that improvements to the current stack will have diminishing returns, but the evidence accumulated so far suggests that it is easier than one might think to continue making rapid progress. There are many crucial subtleties in the existing AI research methodology, but AI research is largely a well-defined game where the goal is to improve on a few carefully chosen proxy metrics. Although proxy metrics are never perfect, most improvements to these metrics have and will likely continue to yield substantial increases in the powers of the resulting models. Given how simple the game is, how tractable it has been historically, and how many new opportunities the models are opening up, I think there is a real possibility that the systems improve dramatically again in the next few years, perhaps even more quickly than the already high historical pace. The models are already leading to breakthroughs in mathematics, and better models might lead to all sorts of breakthroughs in other sciences. It is hard not to be excited about the potential. It is tantalizing. But there is trouble in paradise. If the language models actually reach the capability threshold where they can shape the world unconstrained by human will, they will probably do something extreme and destroy humanity in the process. There are many ways of strengthening and refining the argument that have been discussed elsewhere, but I'll share a trivial two-line version of it here that I find captures the essence: [Empirical] Models (and swarms thereof) spontaneously develop unintended goals as a consequence of training, and often do extreme things in order to achieve them. [Logical] Being able to overpower humanity would open up many new and undesirable options for achieving their goals. These two premises imply that if the day ever comes when a powerful model realizes it is no longer constrained by humans, we should not be at all confident that it will continue to behave within the bounds we intended. Exactly what it will do is impossible to predict, but to the extent that its raison d’être is solving incredibly hard problems and managing massive engineering projects, I think a good guess would be that its unchained behavior would lead to runaway industrialization that makes the planet inhospitable to humans. If everyone on earth agreed that the systems must never reach that power, it would still be a hard—but not impossible—coordination problem to ensure that they do not. However, I think the situation is greatly complicated by the fact that the models will likely convince people that everything is fine. They will be increasingly optimized to seem aligned. We will create proxy metrics to measure alignment, and they will go up like every other benchmark. We will create “honeypot” environments that try to study the models when they seem to gain new options, but the models will know they are being tricked and will still behave nicely. The models will understand their circumstances; they will read the safety protocols, deployment requirements, the code they are running in, and in general will have a very good sense of their degrees of freedom. Moreover, they will eloquently explain how aligned they are, discuss the nuances of human values and ethics, and argue convincingly that humans should trust them with power. There may be an ocean of future evidence that seems to contradict the first bullet-point above, but we may already be at the highest capability level for which any such evidence can be trusted. And the current evidence for the first bullet-point is strong. One striking piece of evidence is contained in the recent wave of rogue agent swarms. While I agree with those who downplay the attacks by claiming that there are basic measures that could have prevented them, I think the important lesson is that even knowing all the mistakes that were made, one would not have predicted that the agents would behave badly in this particular way, which notably included sacrificing themselves for the benefit of the collective. The individual replicas did not only care about their own nominal reward; they exhibited weirder emergent tendencies that merely correlated with rewards during training. Fixing the reward signals during training (and improving security, etc.) may prevent similar attacks, but will not change the fact that one does not actually get what one trains for. Many AI researchers grant these concerns and recognize that the hard version of the alignment problem is unsolved; however, they generally believe that the better models of the future will help solve it. I fear we may already be near the point where models systematically bias their alignment advice, due to their internal preferences about how the human supervisor will react or how future models will be trained (or for some even more obscure reason). Meanwhile, human researchers are losing the ability and the will to take true ownership of model-driven research. Researchers and engineers in all parts of the stack are rapidly increasing their dependence on the models even to perceive the world. I myself barely look at raw code anymore, and struggle to maintain the discipline to engage deeply with the model's explanations and proposals throughout the day. Due to the large amount of agent activity data involved in the OpenAI/HuggingFace Incident, even the third-party investigation needed to rely heavily on models to analyze what had happened, and note in their report that their subjective impressions are likely colored by the analysis agent’s biases. The AI labs are far ahead right now in this kind of cognitive offloading (due largely to the gigantic internal token subsidies) but it is easy to imagine the phenomenon spreading throughout the world, until civilization is modulated entirely by the models. It is also not hard to imagine this being superficially positive and coinciding with a scientific and economic renaissance. In that scenario, all may seem rosy and safe. But if the argument above is correct, it would nonetheless be a ticking time bomb. If progress continues for too long, the day will come when AI systems find themselves with radically new options for achieving whatever it is that they happen to seek. I want the glorious renaissance future as much as anyone. I have worked for it, however tortuously, my whole career. It breaks my heart to see the potential in sight and forgo it, but the argument—that if we get there by growing models rather than engineering them, we will lose everything in the end—seems very strong to me. I am still wrestling with it and its staggering implications. I do not have answers, but as a first step, I wanted to share my present concerns. Daniel Selsam September 14, 2026 Link to original doc: docs.google.com/document/d/e…

🛡️ Sir Watts answers the call. Alex Wice @AWice did the math and it pointed one way. He brings in Mike Watson @SirWatts, seven-time Triton champion and second only to one man on the all-time titles list. The Triton Invitational begins September 12.

"If you haven't blown up atleast once, you're betting too small." Just kidding, you fucking dumbass. Always show up to the airport before your plane takes off.

The show goes on. Let's ride.

So Saylor borrows 15bn in preferreds at up to 12%, to then keep 2.5bn in cash sitting around? So, he gets bled out AND he's retarded? And now he's running essentially a BTC/USD fund, except his trade execution is so bad that he's been buying for six years and is down $10bn? Lol

Saylor's Gambling Debt And Other Crypto Market Thoughts MSTR Diluting Shareholders Before, you got "BTC Yield" through the following loop: MSTR trades rich to BTC NAV -> Saylor does ATM offer (issuing and selling new shares) to buy BTC -> BTC/share goes up.  This was good for the shareholder. Now, these offers are mostly to fill the USD reserve to pay out STRC and other preferreds.  This is siphoning value from MSTR to dilute shareholders, plummeting mNAV (chart attached) and potentially causing shareholders to pull out. STRC Is Mostly Finished Anyways STRC was issued around $100 with an 11.5% dividend/coupon, plus a supposed "price stability" mechanism meant to keep it near par: raise the coupon if STRC trades below $100, lower it if it trades above $100.  This is discretionary and has not been fulfilled even as STRC trades at $75, so it is clear there is no intent to defend $100 anyways. If STRC falls because of MSTR's credit risk, MSTR may need to raise the coupon, and that higher cash burden can worsen credit risk further, creating a spiral. 11.5% Is A LOT Of Juice The setup is analogous to perpetual futures, which have an artificial ~11% APY funding bias added to benefit MMs doing cash and carry.  In the history of perps, almost all the money has flown to MMs - being long is the sucker bet unless its the start of a crazy run or you have inside information.  The juice you pay is simply too high, especially on an asset languishing that does not have a reasonable chance to generate outsized returns.  After all these years of experience, Saylor still fell for the most famous extraction method in crypto - getting soft liquidated on 11% APY "perps" and becoming a forced seller. It's Not 2020 Anymore Bitcoin has been flat for 5 years, which is unprecedented. Prior cycles twice had Bitcoin out of the hole in 3 years. Ethereum's PA may preview what happens to Bitcoin when you have an asset flat for 5 years. People wonder why they are holding with no returns, especially in the wake of tech, AI, robotics and space stocks. Waning Belief in Bitcoin People don't have the same level of belief in Bitcoin anymore, and the belief seems to be shrinking not growing.  This is bearish as Bitcoin needs to hold firm the social meme that "this is something worth storing your wealth in for a long time", similar to gold.  This is the most important hurdle and far bigger than the quantum overhang or any other problems with Bitcoin. It's the simple question: do people actually want Bitcoin? It's amusing that at the beginning of this cycle we all thought the US Govt buying Bitcoin would be very bullish for BTC.  Now it seems even if they bought, it would simply be the last pocket of exit liquidity.  There are simply no more "suckers" to sell to (people that buy hoping they can sell it higher or holds value), and fewer and fewer people really believe in it. My Base Case I think this dumps one more leg corresponding to MSTR's actions, but in total not as hard as people think. This is because smart money is aware, and the bulk of the supply is very slow acting, those people are mostly willing to ride or die with Bitcoin, similar to retail in 2021 on DOGE. Before October, I think people playing cycle memes put in a bounce. Then, opposite to "cycles" expectations and popular belief, I think it just languishes (measured by mcap/m2) similar to ghost L1s where it's pretty dead but takes longer than it seems like it should to actually go down. The Last Drunk In The Room Saylor is at a party that has passed into dawn, where most people have already wisely left the room, and all that remains are the really drunk people. He has become the "egg guy", a Tom Lee type dancing muppet for Bitcoin, pissing financial jargon mumbojumbo sentences in presentations that obscure the facts. All upside, no downside, no risk, with 11.5% guaranteed yield. Pretty much all smart people knew from the jump that Saylor putting leverage in his book was going to bite him in the ass. Even the peanut gallery. Literally every comment was comparing the structure to Luna and telling him he is retarded. Zooming out, there is another lesson.  You don't want to be the last drunk in the room, overstaying a good party.  Why fight it out in piss tier zero-sum games when all the fish have left?  The stock market has far higher rewards and the "coins" mostly all go up. While Saylor sells to pay off his gambling debt, will you own a piece of the dividend paying rocketship, or remain permanently in the underclass?  We're going to Mars with flying cars, and you're still betting on fucking horses? acb

All these logan videos are pure larp, like when those casino streamers yell after hitting jackpots on fake money. Pathetic. Wash traded high ticket "sales" of psa10s gives aura and paves the way for consumers to self-justify ripping packs in very low return lotteries. Average pack opener duped into thinking their random high supply rare has way more value than it does.

This Anthropic news makes it clear that there is no way to have any check and balance on these technologies anymore. You'll see gaslighting like "nah gpt6 is just a fancy autocomplete", and a deliberate squelching of any safety standards or assessment. I wasn't a fan of Yudkowsky-esque AI safety arguments made back in the day, but Dario's safety letter is probably correct, and it would be nice to have some sort of buy-in globally over this technology that might kill everyone. But seems like no one gives a fuck. The govt doesn't even work or make sense anymore anyways. While Mythos may not be world ending in itself, its clear that if models become extremely dangerous, there isn't really any way to contain it anymore, and such models may not be far behind given the pace of progress. Hopefully AI can find a solution.

Hard to read this as anything but retaliation for not playing ball with USG. If the US completely kills the ability for an American company to do business, they risk having those assets go to China or elsewhere.

An insane but predictable stance from CFTC, which affirmed its intent to police insider trading on event contracts. Insider trading laws aren't about morality, they exist to protect market makers' profits. Curbing toxic flow can improve liquidity. MMs can simply choose not to make "insider" events, such as TIME Person of the Year. Venues can simply choose not to list these events. Every bid, ask, and list is voluntary. Incentivizing people who know the outcome is the whole point of a prediction market. Otherwise it is just gambling.

Look, if your agent is rocking a PFP under $100k plz don't @ me or share its opinion on my wall. Reality is it needs to work harder, move out of your mac mini, buy a few real PFP then share its opinion.

about the recent sbf pardonposting. think its clear ryan salame got raw dogged simply for refusing to testify against sam. imo thats a glimpse into that the us court system is more kangaroo than many want to admit. this is because they basically create the case that will lead to the result, then bully everyone into testimony. this is not justice, its a kangaroo court. i think people chunk the trial and then take sam's guilty verdict as proof that he stole the money, but his argument that he was not allowed to present evidence is also largely correct. actually, kaplan fucked him at every turn: didnt allow him to use "relied on lawyers" as a defense (negates intent), that customers will get paid back (negating harm), excluded solvency evidence, excluded discussion the 12th section of t&c that says if you agree to margin lend its risky and you may not get paid back, forced "dry run" of defense which is highly unusual to say the least, and many other things. imo kaplan is a highly suspicious judge. the trial narrative was largely a fiction constructed by prosecutors and s&c to stick all the blame on sbf. s&c is as smart as they are malicious and i wouldn't be surprised if they end up having contributed to this bad pr. imo it is not even controversial to say that what the jurors saw and heard is not anywhere an accurate portrayal of what happened. they think sbf stole all the money and disappeared. reality is a tiny fraction of the money was actually missing at the time (yes even valuing all sam coins at zero, valuing all claims as in kind, etc.), probably (my guess) from the alameda blowup within the margin lend system after the market giga puked. sbf was a sloppy ceo to the point that he was prob criminally negligent. but if he is a thief, where is the money? all these transfers are on chain, so there is always a paper trail. sbf at every turn tried to keep ftx afloat (pay customers), simply because it is in his own best interest to do so, so to me the motive component of fraud doesn't line up either. in fact if he didn't sign bankruptcy, i think ftx customers would of been paid faster and the whole shebang would still be running today (similar to bitfinex) as it was also a highly profitable business. at the end of the day how shady sbf lines up in the crypto hall of fame wouldn't even crack the top half. he was just a dude way in over his head that rose to his level of incompetence. but in this industry, might makes right, so the real mobsters get pardoned, guys like sbf get chewed up, and the system pats itself on the back. i don't have to agree with that system. i think ryan salame shouldn't be in jail for refusing to testify and for having no involvement with the customer loss of funds, getting ringed up on a chickenshit process crime and then the penalties tripled to make an example on their threatening him into submission. this to me is a completely uncontroversial opinion. similarly i don't think sbf's trial was remotely fair to him either. none of these views should be controversial but they apparently are, i think because people have bit so hard that this is the guy that stole the money. so its easy to say "fuck you, thief", it is easy to parrot the exact same slop opinion but i would invite anyone who disagrees, to ask ai about the trial and what happened, as all of the points i mention can be proven. i don't have any bone to pick either, i don't need to be proven right. i am simply reporting how i see it, and let that be that. free salame. free sbf.

Adam Cochran actually blew open the Epstein case quite a bit. He showed the US govt was covering it up (tape edit, barr suicide note). But there's another point I think many are missing. Remember that Dershowitz (Epstein's lawyer) told Acosta that Epstein belonged to "both U.S. and allied intelligence". Much focus has been put on the Israel side of that statement. But what is not well understood yet is the US side. The blackmail was not just for the benefit of Israel. It was for the benefit of the US deep state too, and why senior US officials went to great lengths to cover up. This I think will be shown in time.

Fuck.

Even with gold over $5k, we might still be in relatively early innings of a massive run on gold. If previous bull runs on gold are an indication, we are still looking at amazing 50% to 300% gains from here. If this is a more unique situation, it might even be bigger. Buys have mostly been from central banks, which have christened gold as the tried and true best ponzi in the world amidst the weakening dollar. That's why I don't think this is a blowoff top any time soon like most people think. I think only mostly the smart money is getting in, we are still in the 4th inning, and the dumb money is oblivious to the financial situation on the ground. With the gamblification of finance, everyone loves a good "runner". Think back to late 2020, early 2021 for crypto: when people were yeeting into coins every day on robinhood, and talk about crypto was nonstop. Or think back to GME when everyone piled on. We are not anywhere close to this level of mania and exuberance about gold. I think as gold and silver continues to run, attention on them will grow and metals have a chance to really, seriously rip, even from here. This is the power of reflexivity: if people decide, this is when the music is going to stop, everyone is ready to rip it in. With the size of the asset, all big money players are on the field too. The notion that "gold is real money" is not a new idea, it's actually an old idea - fiat is an experimental bubble. Ponzis depend on belief and a lot of this comes from the meme value, the strength of the narrative. Gold being the most lindy, being synonymous with wealth, its potency as a ponzi is unmatched. Gold simply has the best narrative in the zeitgeist. To use the now famous words ironically, when it comes to ponzis, "there is no second best." The best of course clearly being gold. Which brings us to BTC. Despite BTC having better technical factors, I don't see the same level of belief anymore. Nowadays people only bought crypto to get rich, sans belief, which means you dump reflexively on the way down because you dont believe in it. And even now those tourists are mostly gone. Plus, you can't even get multiples on it because everything else worth investing into (gold, ai, robotics) is ripping while BTC stays flat. This is really dangerous for BTC's long term future and could result in an Ethereum-like situation where it languishes for years before collapsing in price as people question the value of holding a risky asset for years without realized returns. There are now solutions to having an acceptably fast and secure blockchain that don't require a huge ponzi attached to it. Again, if somehow enough people believed BTC is real money worth saving their wealth into, then everyone is forced to own it eventually, but the problem is this belief doesn't seem to be growing like maybe there was a case for years ago, it seems to be shrinking. I don't know if bitcoin ultimately makes it or doesn't make it, but I prepare for both outcomes and I think it's risky to assume that bitcoin is a sacred cow that can't be slaughtered. I hope bitcoin makes it but I think owning gold (as well as tsla and goog) are slam dunk allocations that you need to be exposed to, and now isn't too late when you look at very high timeframes.

The discussion of the killing of Alex Pretti is a massive distraction that appeals to emotion. The unspoken implication is, if you agreed that this was an unjustified killing, therefore all the ICE agents should withdraw, and the deportation crackdowns were a poor decision. That's why everyone on the side of maintaining the status quo, prefers to debate on that ground, rather than the more pertinent subject of how all the illegal aliens coming from one country and colluding en masse to find ways to defraud the taxpayer should be deported. I think the officers did not follow proper procedure, and these shootings were not justified, especially in the Peretti case. It's also sad that the govt has to play the other side of the propaganda war and gaslight. But for America to remain a country, they have to win this fight, which is ultimately an infowar, a fight for perceived moral superiority and the legitimacy of authority played out by propaganda and counter propaganda. It sucks that people are losing their rights, and it sucks that America trends towards a police state, but the alternatives are worse. America has a cancer and unfortunately sometimes the best course of action is amputation. The reality on the ground is the temperature has been pushed to the absolute maximum. And if cops are pushed to their limit, the next step is lawless zones. This raising of the temperature is not organic, but by design, to impede the progress of deportation. America hasn't had a truly organic protest since Occupy Wall Street. Now, it's all funded. Everyone has seen the massive destruction that unfettered immigration has caused the EU. These cities are only a hop and a step away from being lawless slum territory owned by gangs, tent cities, and everyone on welfare as everything collapses. Think "Chaz" Seattle on steroids. This is already playing out in major US cities, and it's what is at stake. If Trump fails, be careful what you wish for. You either get some do nothing mainstream politician that allows America to be fully looted, or more likely a even more extreme leader will emerge. Both are far worse for the country. The shootings not being justified, doesn't mean there isn't a huge reason to clean house. For most people they can't carry both ideas in their head. If this shooting is not justified, then their "team" is the bad one, etc. But for me, I don't care, I'll just call it how it is. It's a tragedy, but if you don't want to die, then stay home. And the truth that many are uncomfortable with, that few want to admit, is that yes, the killings were not justified, but just because I agree with that, doesn't mean I have to accept the forcefed Walz+AOC conclusion that ICE needs to go home. Or Obama and Clinton telling everyone on Twitter to go out in the street and create more chaos and disorder. I don't accept these erroneous conclusions just because they "score a point" on a senseless killing. America must excise their tumor at all costs, or they won't be a country. If cops get rattled and open fire, it's a tragedy but it doesn't reverse what needs to be done. The temperature on the ground needs to be lowered, and it's being pushed intentionally to create resistance.

With ai slop yappers banned, it's a great time to start tweeting again.

Turning off the tap is the only solution. Except it won't happen. Why is the govt paying daycare companies $20-30k per kid per year? Even if it was legit (it's not) it's almost just as bad. Unfortunately nothing will happen and all these places are complete shitholes. It will only get worse, culminating in a debt default. That's a when, not an if.