"NO ONE KNOWS HOW LONG THE SHORTAGE WILL LAST" - FULL $SKHY CEO INTERVIEW SK Hynix CEO Kwak No-jung predicted that the memory semiconductor supply shortage will continue until the end of 2030, and that there will not be a sharp decline in demand thereafter. Regarding plans for additional investment in the U.S., he reaffirmed the existing stance that investment is possible anywhere sufficient support is provided. CEO Kwak gave this response to a question regarding how long he expects the memory supply shortage to continue during a press conference following the groundbreaking ceremony for an advanced packaging production base for AI memory at Purdue University’s Holloway Gymnasium in West Lafayette, Indiana, on the 27th (local time). "No one knows how long the shortage will last, but there are no clear signs of a downturn," said CEO Kwak. "We expect the shortage to continue until the end of 2030." He further explained, "I believe the next downturn will be different from the ones we have experienced over the past few decades. While adjusting production volumes appropriately is not easy, our business model for AI is transitioning from 100% standard products to partially customized or fully customized products." In the case of AI memory, since it is custom-made for client systems, demand forecasting is easier and ties with manufacturers are stronger compared to the past. Consequently, the explanation is that even if a downturn occurs, the damage will not be as severe as in the past." Even if a downturn hits, I believe demand will slow down slightly or remain at a stable level rather than plummeting," CEO Kwak said. "Therefore, I am not concerned about the period after 2030. I believe supply and demand will balance out, and the entire AI industry will grow together." When asked about plans to invest in additional semiconductor production facilities in the U.S., he replied, "We are open to any location where sufficient resources, such as water, electricity, talent, and subsidies, are provided," adding, "Since nothing has been finalized yet, I cannot discuss specific details or directions." Last month, SK Group Chairman Chey Tae-won stated, "If there is a suitable place, it does not matter whether it is the United States or anywhere in the world," expressing the same position. However, CEO Kwak specifically added, "I hope to be able to deliver good news to the U.S. AI industry next time." When asked if he felt pressured by the Donald Trump administration's demand for investment in memory semiconductor front-end process facilities, he replied, "Our current situation is very well known through media reports." He reiterated, "We are reviewing any location in the world as long as it offers customers and new business opportunities," adding, "If the conditions presented by the other party are met, we would like to deliver another piece of good news to the AI industry in the U.S. or other countries, just like at today's event." Regarding whether to continue maintaining its stake in the Japanese NAND flash company Kioxia (formerly Toshiba Memory), he answered, "There are no concrete or confirmed plans yet." He further explained, "However, just as we contribute to the U.S. AI industry, we also intend to contribute to the Japanese semiconductor industry. Since we are investing in Japan just as we have in the U.S., we view it as our responsibility in some respects to contribute to that local community and the nation. Although there are no confirmed plans, we are carefully examining how we can grow the market together." Source: newsis $MU $SNDK $KXIAY

full bull, but with intention and experience Onchain bull run in an emerging social app trading cycle, all while bitcoin is still trading below 80k in an uptrend and rh chain is just two months old. BUT, I know it's not what traders want to read when they're trying to make a lot of money in a short amount of time, but don't forget to TP and offramp gradually, especially when you're euphoric. Let it be $500 a week or more, doesn't matter, it's your journey. More opportunities will come, yes, but comfortable trading size can be an edge against emotional overtrading. That's how I made it in 2024, by actually off ramping capital to my bank accounts and protecting it by doing nothing. For most traders, wealth builds over time, not with one big trade that makes you a fast million. I still have all my 2024 trading profits comfortably sitting in an ETF, not left onchain to slowly bleed out in a bear market, bc capital sitting onchain has a way of getting used for "just one more trade." CT pushed the narrative back in 2020 that meme stock szn was your last chance to make it. Then altcoin szn in 2021 was the last chance. Then meme coin szn in 2024 was, of course, your last chance, and AI coin szn in 2025 was the last chance too. And now it's supposedly the last chance to make it with the rh eco. There will always be a next "last chance." So stay calm, have a trading framework, and stay true to your principles, but also take the risk sometimes and have the balls to hold a little longer, especially in an emerging bull market like right now. A good trader with a proven edge always makes money, bc he's consistently profitable in every market and dynamic. Don't let your trading framework get broken by emotional decision making, sizing up against your own rule set. Be smarter, especially while watching high uPnL like this on fomo. You can watch me trade this bull market, offramp my capital, and do daily thesis work, all transparently, on fomo. ref link:

some thoughts on Arc chain: ppl building on Arc or related to Arc have reached out to me several times over the past few weeks apparently, they want to get trenchers onto the chain early because they understand that memecoin activity is one of the fastest ways to bootstrap traction for a new chain i mostly ignored them i've been too busy to even keep track of all the messages i get these days, but more importantly, i've believed that onchain activity this cycle will be concentrated around Robinhood, BNB and Solana that was my bias until i learned that fomo will be integrating Arc shortly after launch Seyong from fomo recently told me that the integration should happen within a few days of launch and honestly, that changed my short-term bias on Arc don't get me wrong i still believe RH/BNB/SOL will be where the majority of the real action happens this cycle but there's something important to consider: fomo hasn't even integrated Hyperliquid, Tron or TON and all three of those chains produced runners that reached $100m+ so if fomo is integrating Arc right out of the gate, i think there's a very real possibility that there will be some serious short-term money-making opportunities for trenchers and that's the key distinction: short term i am NOT saying Arc is going to become the next BNB, Robinhood or Solana i'm saying this is the first time fomo will be integrating a non-mainstream chain essentially from launch that makes Arc a very interesting experiment and i'm positioned accordingly i bought the native token of the main launchpad on Arc, $LONG, along with the top 3 memecoins on the launchpad my thesis is basically this: 1. fomo is integrating Arc fomo is becoming one of the main crypto trading apps for retail this cycle i originally ignored Arc entirely the fomo integration changed that 2. Arc is actively trying to attract trenchers multiple ppl building around Arc have reached out to top trenchers like myself that suggests they understand how important early memecoin activity can be for a new chain 3. they appear to be trying the Robinhood playbook i've also had a source tell me that Arc understands how important it is to have trenchers on the chain early and is actively focused on making that happen essentially, they're borrowing a page from the Robinhood playbook and i want to be super, super clear about this: this is a degen, degen rabbit hole this is a short-term adventure i'm not rotating my core thesis away from RH/BNB/SOL and i genuinely believe that RH/BNB/SOL is where you want to concentrate your energy this cycle... and that most other chains will likely be a waste of time on a long enough timeframe so why am i touching Arc? i'm simply betting that the combination of new chain + launchpad + trenchers + fomo integration could create some very interesting early opportunities if you're interested in exploring it, you can either wait for the fomo integration later this week or bridge over earlier through the main launchpad on Arc, Long for those who want to bridge: 1. buy $CRCL on Robinhood (most tokens on Arc are paired against $CRCL): 2. bridge your Robinhood Chain $CRCL to Arc using the bridge provided by the main launchpad: 3. buy whatever you want through IMPORTANT: don't use trading bots like Basedbot or GMGN right now they're routing through the wrong pools and you could end up losing a significant amount of your money to slippage and if you're not on fomo yet, you really should hurry to get your account set up if this thesis plays out, you don't want to be setting everything up after the opportunities have already appeared join through my ref: i'm also doing a $350k giveaway for active fomo refs so you could literally get rewarded for being early again: this is a short-term degen experiment but if you're a trencher, this is exactly the kind of setup i'd want on my radar before the masses arrive

📌Alice’s One-Liner Principle Vol. 4 💡Better Odds Don't Mean Better Chances. A lot of people watch odds slide from 2.0 to 1.8 and immediately think something big has happened. The market is moving, so there must be information they do not have, and they rush to follow. Alice wants to remind you that a move in the odds is a snapshot of market mood, not an update of the facts.📈 💎Take this example. In a Bundesliga match, Dortmund went away to Werder Bremen. In the 24 hours before kickoff, Dortmund’s odds kept falling from 1.85 to 1.62 as heavy money piled into the away side. It looked like a lock. ❤️‍🔥What happened? Dortmund drew 1-1. The odds kept dropping and the game still did not get won. It was not because someone had inside information. Markets can misread a situation together. Money sometimes floods one way only because the story looks too tidy, not because that story is actually going to happen. Odds tell you what other people are buying. They do not tell you what the match will do. 💫Alice’s advice: Next time you see a clear move in the odds, ask yourself: 1⃣Is this move tied to a specific event, such as a key player returning, or is it just money chasing money? 2⃣If the odds had not moved at all, what would my own judgment still be? The market can show you where the money went. Where the money went does not mean the result will go there too.👀 🧐What scenes have you seen where the odds kept falling and the team still failed to win? 👇 #PredictionMarket# #TradingTips# #AliceAI#

hey fam, good morning. hey @fomo @seyong @patagucci_girl @renyukong @qrazhan_ultd here's two important builder requests from us @prismassets : 1) cross app support: we have a staking feature where $prism token holders can stake their bags to get staking rewards but as most of our holders are on fomo too, they are unable to stake it. can we get cross app support from fomo to embed staking for fomo wallets? i believe there will be hundreds of protocols in the future which offer staking and fomo being the dominant social trading app, it's really necessary too. 2) token verification: robinhood:0x20024e485c0b22b42855589700721b28320a7777 is trading above 3m mcap for a long time with daily $1m volume multiple times and i have applied for fomo token verification multiple times but somehow we are still not verified. ca: 0x20024e485c0b22b42855589700721b28320a7777 i tried every possible way to verify it, still i'm unable to proceed with it. please help us with verification: we are currently at $4.2m mcap. please help us with this as it's long pending. today we will work on: - rwa perps and social trading layer for launch - work on v2 staking pool with better cooldown and more emissions - working on improving dex emissions and adding more pools - finetune the protocol more for much smoother onboarding - add more cool assets - work on partnerships - building the agentic marketplace in collaboration with multiple good builders some good people in my community told me that rwa prediction markets can be a cool idea to integrate and would attract more people. i'm still researching on it and will let you guys know soon. we are also working on an installable mobile web app which should go live next week. there's a lot of stuff that needs to be worked upon and a lot of issues to be resolved this week as we are currently working with some really good people for protocol growth. whatever, we keep shipping. let's build for 16 hours straight today, together letsgooooo

I don’t think people fully understand what Tesla is about to show us on October 1... this might be the most exciting Tesla event we’ve ever seen. And that’s not just me trying to hype it up. Elon himself has said he feels confident this will be “the most exciting product unveil ever,” potentially “one of the most spectacular demos ever,” and has described the Roadster as the coolest car ever made… “if it even is a car.” We could be watching Tesla blur the line between a car, a rocket and something we’ve never seen before. I really can't wait... here’s everything Elon has said about the next-gen Roadster, word for word, ranked from the craziest to the least crazy: 👇👀 1/ “The new Tesla Roadster can fly” — Elon on 𝕏, reply to Marc Andreessen (Jun 15, 2024) 2/ “Tonight, we radically increased the design goals for the new Tesla Roadster. There will never be another car like this, if you could even call it a car. … 0-60mph < 1 sec. And that is the least interesting part.” — Elon on 𝕏 (Feb 28, 2024) 3/ “SpaceX option package for new Tesla Roadster will include ~10 small rocket thrusters arranged seamlessly around car. These rocket engines dramatically improve acceleration, top speed, braking & cornering. Maybe they will even allow a Tesla to fly …” — Elon on 𝕏 (Jun 9, 2018) 4/ “I want it to hover. I’m trying to figure out how to make this thing hover, without killing people. … maybe it could hover, like, a meter above the ground… Maybe, I don’t know, six feet. [Can it travel while hovering?] Yeah. At a minimum, I’m confident we could do a thruster where the license plate flips down, James Bond-style… 3 tons of thrust… We’re going to throw some rocket technology in that car.” — Elon on Joe Rogan Experience #1609# (Feb 10–11, 2021) 5/ “We’re gonna use ultra high pressure compressed air. It’s a cold gas thruster. The main thruster will be behind the license plate. So for acceleration, it drops the license plate, and behind the license plate is a rocket thruster. It’s like full-on James Bond. It’s going to have an option to have rocket thrusters from SpaceX. No, in this case I am serious.” — Elon on Jay Leno’s Garage (May 22, 2020) 6/ “Yes, with the SpaceX rocket thruster option package. It will be safe, but very intense. Probably not wise for those with a weak heart or vestibular system.” — Elon on X, reply on 1.1-second 0–60 with SpaceX package (May 20, 2021) 7/ “The new Tesla Roadster will be the fastest production car ever made, period. And this is the base model. 0–60 in 1.9 seconds. 0–100 mph in 4.2 seconds. There’s only one thing that’s beyond ludicrous… Plaid.” — Elon at the Tesla Semi & Roadster Unveil in Hawthorne (Nov 16, 2017) 8/ “This will actually be able to fly very briefly. I always laughed at flying cars & now making one. Fate 🖤 irony haha.” — Elon on 𝕏 (May 21, 2021) 9/ “Under 1 second. It does not have big wings. It will have a drive-by-wire yoke, essentially. Like an aircraft or like how modern jets are controlled.” — Elon in the Don Lemon interview (Mar 18, 2024) 10/ “We’re going to put rocket thrusters on it. Nothing. Ultra high pressure compressed air. Cold gas thrusters. Yeah, it’s an electric pump. Pump it up to like 10,000 PSI. We could make it fly. Yeah, 600 miles. We could totally make one right now that would do 600 miles. But the thing is, it’s too expensive… just have a 200-kilowatt-hour battery pack.” — Elon on Joe Rogan Experience #1169# (Sep 6, 2018) 11/ “Yes. Not recommended for urban environments. Exactly. Total energy stored even in ultra compressed air is low vs battery, but power output is insane. The composite overwrapped pressure vessel (COPV) is most advanced ever made. It’s what SpaceX is qualifying for NASA crewed missions. Extremely robust — literally bulletproof. Trade is just that Roadster becomes a 2 seater instead of a 2+2.” — Elon on 𝕏 (Jun 10, 2018) 12/ “I’m not. Will use SpaceX cold gas thruster system with ultra high pressure air in a composite over-wrapped pressure vessel in place of the 2 rear seats.” — Elon on X, reply to MKBHD (Jan 9, 2019) 13/ “If safety is your number one goal, do not buy the Roadster… We’ll aspire not to kill anyone in this car. It’ll be the best of the last of the human-driven cars. The best of the last.” — Elon on Moonshots with Peter Diamandis (Jan 6, 2026) 14/ “In fact, long term, the only manually driven car will be the new Tesla Roadster. Speaking of which, we may be able to debut that in a month or so. … I think it will be one of the most exciting product unveils ever. … I think it might be one of the most spectacular demos ever.” — Elon on the Tesla Q1 2026 earnings call (Apr 22, 2026) 15/ “Yes. Eventually. We’re getting close to demonstrating the prototype. … one thing I can guarantee is that this product demo will be unforgettable. Unforgettable. Whether it’s good or bad, it will be unforgettable. If Peter wants a flying car, he should be able to buy one. I can’t do the unveil before the unveil. I think it has a shot at being the most memorable product unveil ever. This is some crazy, crazy technology we got in this car.” — Elon on Joe Rogan Experience #2404# (Oct 31, 2025) 16/ “I feel confident in saying it will be the most exciting product unveil ever. … the new Roadster is very much … not even the icing on the cake, it’s the cherry on the icing on the cake. … It’s the coolest car, if it even is a car, that has ever – that will probably ever exist.” — Elon at the Tesla 2025 Annual Shareholder Meeting in Austin (Nov 6, 2025) 17/ “Roadster’s kind of like dessert. … I think we should do Cybertruck first before Roadster. With the Roadster, we’re going to do some things that are kind of unfair. … from kind of like Rocket World and put them on a car. When we do the unveil of the Roadster, let me just say that anyone who’s been waiting, they won’t be sorry. That’s the base model. … Faster than that. Let’s just say faster than that.” — Elon on Joe Rogan Experience #1470# (May 7, 2020) 18/ “Intent of new Tesla Roadster is to beat gas sports cars on every performance metric by far, no exceptions, thus transferring the “halo crown” effect gas cars have as the top speed leaders over to pure electric.” — Elon on 𝕏 (Jun 10, 2018) 19/ “You have my word, Marques. Next gen Roadster will be absolutely out of this world. For those who love to drive, there is no finer car in history and we don’t think there will be another.” — Elon on 𝕏 (Jun 10, 2018) 20/ “2.1 sec 0-60 mph is base model before adding rocket thruster option” — Elon on 𝕏 (Jun 27, 2019) 21/ “SpaceX thruster package will be subtle. Hidden behind license plate.” — Elon on X (Jul 13, 2019) 22/ “New Roadster is part rocket” — Elon on 𝕏 (Jan 28, 2021) 23/ “Not saying the next gen Roadster special upgrade package will definitely enable it to fly short hops, but maybe … Certainly possible. Just a question of safety. Rocket tech applied to a car opens up revolutionary possibilities.” — Elon on X (Nov 19, 2017) 24/ “Should clarify that this is the base model performance. There will be a special option package that takes it to the next level.” — Elon on 𝕏 (Nov 19, 2017) 25/ “New Tesla Roadster will exceed all gas sports cars in every way…” — Elon on 𝕏 (Jun 9, 2018) 26/ “Oh sorry, yes, Roadster will have thrusters, not Plaid. Plaid is indeed close to the theoretical maximum of a roadworthy car with rubber tires.” — Elon on 𝕏 (May 20, 2021) 27/ “The Roadster is not just the icing on the cake, it’s the cherry on the icing on the cake. … we’d all love to work on the next-gen Tesla Roadster, it is super fun, and we are working on it – but it has to come behind the things that have a more serious impact on the good of the world. … We’re close to finalizing the design on that, it’s truly gonna be something spectacular. … Peter was lamenting on how the future doesn’t have flying cars. Well, we’ll see.” — Elon on the Tesla Q3 2024 earnings call (Oct 23, 2024) 28/ “Where is that thing? We expect to complete the engineering and design of the next gen Tesla Roadster this year. Hopefully start production, this is not a commitment, but hopefully start production next year. … it’s not even the icing on the cake, it’s the cherry on the icing on the cake. … the next gen Roadster will have the SpaceX option package. That’ll make it truly next level.” — Elon at the Tesla 2023 Annual Shareholder Meeting (May 16, 2023) 29/ “I really think long-term, the only vehicles that we’ll make will be autonomous vehicles with the exception of the next generation Roadster which we’re hoping to debut in April. It’s gonna be something out of this world. Fantastic.” — Elon on the Tesla Q4 2025 earnings call (Jan 28, 2026) 30/ “The new Roadster is something special beyond a car” — Elon on X (Sep 17, 2025) 31/ “New Tesla Roadster Unveil 10.01” — Elon on 𝕏, quote-tweet of Tesla’s “Go for launch” (Sep 13, 2026) 32/ “Next gen Roadster will be convertible” — Elon on 𝕏 (Apr 13, 2017) 33/ “Note, gas contained would be ultra high pressure air in a SpaceX rocket COPV bottle. The air exiting the thrusters would immediately be replenished whenever vehicle pack power draw allowed operation of the air pump, which is most of the time.” — Elon on 𝕏 (Jun 10, 2018) 34/ “Using the config you describe, plus an electric pump to replenish air in COPV, when car power draw drops below max pack power output, makes sense. But we are going to go a lot further.” — Elon on 𝕏 (Jun 9, 2018) 35/ “The new Roadster will actually do something like this” — Elon on X, quote-tweeting a hovering DeLorean GIF (Jan 9, 2019) 36/ “Yes. To be clear, Plaid powertrain is about a year away from production & applies to S,X & Roadster, but not 3 or Y.” — Elon on 𝕏 (Sep 12, 2019) 37/ “My favorite is one in tension, other DoF in compression. New Roadster wheels will only have one nut.” — Elon on 𝕏 (Aug 18, 2020) 38/ “It would also obviously be ridiculous for the Roadster to achieve simply the same 0-60mph time as Model S, even though it is a 2 door fully performance-oriented vehicle” — Elon on 𝕏 (May 13, 2024) 39/ “We have a few ideas for increasing torque & max rpm even further for new Roadster. Definitely fun & exciting engineering ahead!” — Elon on 𝕏 (Jun 13, 2021) 40/ “Yes. Model S will always be the fastest Tesla until next gen Roadster, which is a few years away.” — Elon on 𝕏 (Mar 24, 2017) 41/ “With respect to Roadster, we’ve completed most of the engineering. And I think there’s still some upgrades we want to make to it, but we expect to be in production with Roadster next year. It will be something special.” — Elon on the Tesla Q2 2024 earnings call (Jul 23, 2024) 44/ “We will have special colors for new Roadster, as we did for the original” — Elon on 𝕏 (Nov 25, 2020) October 1, 2026 is going to be the next-gen Tesla Roadster Unveil... I hope everyone is mentally and physically ready.

BILL GATES WHERE THE ART THOU?! I hope you're watching the Tesla Semi event tonight... Tesla opens the Semi factory tonight at 6pm PT, live on 𝕏. In 2020 Bill wrote that an electric 18-wheeler would probably NEVER work. "The problem is that batteries are big and heavy. The more weight you're trying to move, the more batteries you need to power the vehicle. But the more batteries you use, the more weight you add, and the more power you need. Even with big breakthroughs in battery technology, electric vehicles will probably NEVER be a practical solution for things like 18-wheelers. Electricity works when you need to cover short distances, but we need a different solution for heavy, long-haul vehicles." BTW, he also bet against the company... he shorted $TSLA. In 2022 Elon texted him and asked if he still had a $ half-billion-dollar short position against Tesla. And he did. On December 17, 2025, Elon said that short was about 1% of Tesla's shares, and that it might have cost Gates over $10 billion by now. Elon even once said Gates told him in person, at Giga Texas, that a long-range Semi was impossible. The Tesla team has been grinding over the years as people gave up on the Tesla Semi program... every year it slipped, more of them walked away. In The Master Plan, Elon wrote the goal down on July 20, 2016... Tesla was NOT going to stop at cars. Heavy trucks had to go electric too, bc that is where a ton of the fuel actually gets burned. The team showed the Semi on November 16, 2017 and PepsiCo took the first ones in December 2022, hauling real freight with them ever since. And tonight they are opening the factory. The building is 1.7 million square feet, next to Gigafactory Nevada. Tesla sized it for up to produce 50,000 Semis a year. And Dan Priestley has said there will be many thousands by the end of 2026. So here are the facts. 1/ Standard Range is about 325 miles. Long Range is about 500. Both of those are at a full 82,000 lb load. 2/ It uses about 1.7 kWh per mile. A diesel Class 8 usually gets around 6 to 7 miles per gallon. A gallon of diesel holds about 38 kWh of energy. At 7 mpg that is a little over 5 kWh of fuel burned every mile. The Semi uses 1.7. The diesel is burning about 3 times the energy to move the same freight. 3/ The long range truck can take up to 1.2 megawatts. Tesla says about 60% of the battery comes back in 30 minutes. That is a driver’s required break, and on the 500-mile truck it is on the order of 300 miles. There is also a 25 kW outlet on the truck, so a refrigerated trailer can run off the battery. 4/ After the driver, fuel is one of the biggest bills a fleet has. $4 diesel at 7 mpg is about 57 cents a mile. Electricity at 15 cents a kWh and 1.7 kWh a mile is about 26 cents a mile. The direction is a lot less $ money per mile, and the price of electricity does not swing around the way diesel does. You also drop the oil changes, the diesel exhaust fluid, and the emissions equipment that keeps parking diesel trucks in the shop. Regenerative breaking slows the truck down, so the brakes last longer. It's clearly a financially better decision for businesses. 5/ Medium and heavy trucks were 23% of US transportation emissions. Transportation was the largest source of America’s direct emissions. These trucks are a small share of the vehicles, yet a huge share of the fuel. Take one diesel doing 100,000 miles a year at 7 mpg. That is about 14,300 gallons. The EPA puts about 22.5 pounds of CO2 in a gallon of diesel. Call it 145 metric tons a year, out of one tailpipe. The Tesla Semi also has no tailpipe, no diesel soot, no NOx on the street next to the warehouse, the port, or the school by the highway. A diesel is stuck burning diesel for the rest of its life. 6/ 800 kW is about 1,070 horsepower, from three motors. A lot of diesel tractors are 400 to 600 horsepower, and a full trailer makes them crawl on a grade while cars stack up behind. The Tesla Semi has the power to keep its speed. Tesla says those three motors can also catch a jackknife. The battery sits low, which makes the truck harder to roll. And Tesla Semis are designed for autonomy. 7/ PepsiCo has been running Semis since December 2022. Einride ordered 500, starting this month, across California, Texas, New Jersey, Illinois, and Georgia, including Amazon freight. This week ZET SCALE, a shipper group with Microsoft and PepsiCo in it, picked Tesla as the primary truck for a program of up to 2,500 electric Class 8 trucks. Tesla’s own account said 2,500 Semis. Kenworth, Volvo, and RIDE are also named in that deal, so the split is one of the things to listen for tonight. IMC ordered 50 for California port runs. DHL and ABF already bought after real pilots. Semis spotted at the factory today already had customer logos on them - PepsiCo, US Foods, DHL, Einride, ABF, IMC, WattEV, OK Produce, and more. So Bill, I hope you’re watching tonight. The truck you said would probably NEVER work is now getting its own 1.7-million-square-foot factory ready to be built at scale. The debate over whether an electric 18-wheeler can exist is over. Now the question is how big this gets. And knowing Tesla, the Semi we see today is probably only the beginning.

for some reason people seem to have Anthropic Derangement Syndrome over a pretty straightforward combination of cryptographic primitives and LLM sampling techniques. so as someone who isn't an employee, let me try to explain what's going on with watermarking step by step. - LLMs work by probabilistically autoregressively sampling tokens. meaning: at every token, the model weights don't output a single next token, but a probability distribution over *all possible tokens*, a bunch of little numbers that sum to 1 - the "temperature" sampling setting affects how this distribution is constructed. at 1, it's just the "ground truth" / whatever the model thinks. shifting it above 1 will make the distribution much more fat-tailed, lower probability tokens will be higher probability etc. shifting it towards 0 makes it closer to deterministic, making the highest probability tokens much more likely to be selected, and at 0 always just selecting the single most probable option - modern reasoning models almost exclusively use temperature 1, and apis often no longer even expose it as a customizable setting, so "deterministic generation" isn't common - when you "pick something randomly" on a computer (not just LLMs), almost always it's actually *pseudorandom*, eg using a complex algorithm that based on an initial seed number, generates a chain of numbers that has nice cryptographically provable properties: the distribution of the numbers has no pattern, no information content, and can't be predicted better than chance by any method except having the initial seed number and in fact running the same pseudorandom algorithm. that means you *can* deterministically generate the same "randomness" if you have the same seed, but no one else can tell the difference between that and real randomness - often these pesudorandom number generators are initially seeded either with something like time, or for more security using a hardware randomness source, something that samples physical temperature or the like on-chip. those sources are too slow and expensive otherwise use for all randomness - pre watermarking, when Anthropic generated Claude tokens it would use pseudorandom generators to select those tokens from the LLM distribution, with the generators seeded in an ~unknown but generic way. very likely just some system default which is one of the above, but this is completely opaque to the end user. as mentioned, *tokens are already selected randomly*, it is not the case that LLMs always pick the most likely next token, that in fact is very undesirable and leads to much lower quality generations - post-watermarking, the only thing that changes is *how the seed is selected*. now, it's always *seeded* with a deterministic hash of the prior tokens plus the secret key. the selection is still *psuedorandom*, with exactly the same properties described above. it's still the case that cryptographically, at each given token, if you look at the full token probability distribution the LLM outputs and see which one the sampler selects, its actual choices are indistinguishable from having used a source of physical randomness *unless you know the seed*. it doesn't cut out certain words from the vocabulary, it doesn't "affect phrasing", any more than the existing system of random selection already does - the only difference is now, it's possible for Anthropic to take a sequence of words, run each token through Claude to get the LLM probabilities per token, then seed the same pseudorandom number generator in the same way with their secret keys, and *check whether the token selected matches the one the pesudorandom generator would have selected when seeded in that way*. the property they're taking advantage of is that PRNGs are in fact deterministic, so if this exact setup was used to generate the tokens then all the selections will match exactly, and this will be wildly wildly implausible / virtually zero probability on any meaningful sequence of words - what this will not do: it won't (and can't) differentiate between very very very overdetermined content. for example if you prompt an LLM with "What is 1 + 1? output nothing besides the numerical integer answer", then the token distribution is: 2 with probability 99.9999%, and then every other token in the universe with negligible probability. no matter how you seed your PRNG (which again, all LLMs already use for sampling), all the probability mass is on 2! the randomness is used for selection weighted by probability, and there aren't any other probable choices here - but it really doesn't require very many token choices for this to come up, because the vast majority of English is *not* overdetermined. you can personally inspect the LLM output distributions in various playgrounds and see how many often quite close to equal probability tokens are sitting near the top. as mentioned these are *already* being selected between, never just selecting the single most probable, so within a sentence or two of normal output there will be overwhelming (but undetectable without the private key) watermark evidence - also, they're using Google's SynthID algorithm for this, and *gemini already does exactly this and has for like a year and a half*. basically all text you've gotten from gemini has already gone through exactly this process! hopefully that clears things up somewhat!

thoughts on coxon 1. he appears to have worked at OpenAI for 3 years and Anthropic for 2 months. Nobody seems to be disputing that 2. There’s a lot of money being spent on amplifying AI safety messaging, as well as anti-AI messaging. YouTubers have spoken publicly about being offered money to make videos etc. Whether Coxon himself was aware of it or not, some of these influence operations quite likely boosted this story along with people generally being interested to hear they might all die. 3. I think Coxon is right that the current incentives are not aligned towards safety. Instead the current incentives are for an arms race. Neither Coxon nor anyone else has proposed a solution to change these incentives. Resigning won’t change anything, making noise in the media itself won’t change anything… even the United States acting unilaterally won’t change anything. This is real and should be concerning, but the probability of human extinction from AI this month is still near 0. 4. Superintelligence is now inevitable. Once we had transistors, microprocessors were an inevitable next step. You could take out Intel, or get them to agree to slow down, but someone would eventually design and manufacture a microprocessor. In the same way, once the ingredients for creating superintelligence exist, someone will create it. 5. Superintelligence will be able to destroy the world, or kill all humans. If it is smart enough to do anything, it is smart enough to kill us if tasked with doing so. It could hack in to our utilities using zero days or social engineering, shut down our critical services, and wreak havoc just as easily as it could vibe code some shitty SaaS app. It might decide to kill us on purpose (if it was tasked to), on accident (if it was tricked into doing so) or inadvertently (just by competing for economic resources so aggressively that goods and services need to sustain human life are crowded out) 6. We have no idea how to build a model that is capable of killing everyone, but won’t do it 7. Even if we do figure out alignment, not every model in the world will be perfectly aligned with human values. Many will be misaligned inadvertently or intentionally. 8. It seems safe to assume that not only will AI get smart enough to kill of humanity if it wanted to, but in fact there will inevitably be a super intelligent AI that tries to kill us 9. Our only hope for survival is that on the day that happens, we have an even smarter model that is well aligned and can defend humanity against this threat 10. Superintelligence is a super weapon, and is the future of warfare. Why send soldiers when you can send intelligent robots? Better yet, why even send robot soldiers when you can just exploit and take down a country’s power grid? Seen from the perspective of avoiding war and death, all of these offensive capabilities can be justified defensively. Billions and trillions will go into honing these models offensive capabilities, but it’s unclear how to ensure those capabilities are only used for good 11. We should probably study the history of the last time we created a super weapon that could destroy the world: the atom bomb. AI is very different but both technologies have the potential to end humanity. People were very afraid at the dawn of the atomic age, and many thought the world would end. A few times we got close to nuclear war. But ultimately we got MAD and decades of relative peace. 12. Concern and open debate is good here. But I believe nobody has proposed a reliable way to slow or stop progress. Given that reality, the most important thing for safety may be to stay ahead of malicious superintelligence, and trying to slow down unilaterally could in fact be extremely dangerous. We need open and honest debate to find the answers and use this technology to help people. Sensationalized media stories almost set the conversation backwards

Jackson hole会议有关ai的摘录 1，AI的发展速度与经济增长潜力 Capital and labor have combined to create the large language models at the heart of AI. Users buy tokens to gain access to the models. Reports put annualized token sales for the two leading labs alone at more than $100 billion—an increase of 500-plus percent from a year ago. 资本和劳动力共同创造了处于AI核心地位的大语言模型。用户通过购买Token来使用这些模型。据报道，仅两家领先实验室的Token年化销售额就已超过1,000亿美元，同比增长逾500%。 2. AI可能成为新的生产要素 The Fed watches all of this attentively. We recognize that AI is a new variable—potentially a new factor of production—that will have consequences for both the economy and the conduct of monetary policy. It opens some major lines of inquiry:美联储正在密切关注这一切。我们认识到，AI是一个新的变量——甚至可能成为一种新的生产要素——它将同时影响经济运行和货币政策的实施。这带来了一些重要的研究问题： Will the application of AI cause a significant, sustained rise in productivity across the economy? And if so, when?AI的应用是否会推动整个经济体的生产率显著且持续提高？如果会，这种变化将在何时出现？ Will token usage be complementary or competitive to labor? Will the next generation of AI models demand even greater capital intensity, or will the models themselves help devise a capital-light solution? Token的使用与劳动力之间究竟是互补关系，还是竞争关系？下一代AI模型是否需要投入更多资本，抑或这些模型本身能够帮助开发出资本需求较低的解决方案？ 3. AI产业的收益将由谁获得 Among the other yet unknowns is the resulting market structure. It's not obvious where the returns on capital will land or on what timescale. Early on, how much of the surplus goes to owners of scarce assets—AI labs, chipmakers, energy producers, and cloud providers? Over time, how much of that value accrues to businesses and consumers? What are the broad implications for workers and for the employment side of the Fed's mandate?另一个尚未确定的问题，是AI最终会形成怎样的市场结构。资本收益最终将流向哪里、又将在多长时间内实现，目前并不明确。在发展初期，经济剩余中有多少会流向稀缺资产的所有者——包括AI实验室、芯片制造商、能源生产商和云服务提供商？随着时间推移，又有多少价值会由一般企业和消费者获得？AI将对劳动者以及美联储就业使命产生哪些广泛影响？ 4. Token价格与模型分层 Likewise, we don't yet know the equilibrium price of the tokens. Might there be a heterogeneity of tokens, such that growing sums will be paid for access to the best models at the frontier? Will token prices for older models fall to the level of their marginal cost?同样，我们尚不知道Token的均衡价格。未来是否会出现明显的Token分层，使用户愿意支付越来越高的费用来使用最先进的前沿模型？较旧模型的Token价格是否会下降至其边际成本水平？ 5 美联储将成立专项工作组 We will be thinking through these matters with the help of a task force on productivity and jobs. My early check-ins with the leaders of that task force, and the four others, have been encouraging.我们将借助一个专门研究生产率与就业问题的工作组，深入分析这些事项。我与该工作组以及另外四个工作组负责人的初步交流令人鼓舞。To be clear, though, their recommendations will come later and have no bearing on decisions we make in the current policy conjuncture. But I believe that for future policy challenges, this intellectual investment today will leave us far better prepared.不过需要明确的是，这些工作组的建议将在以后提出，不会影响我们在当前政策环境下所作的决定。但我相信，今天投入的这些研究工作，将使我们在应对未来政策挑战时准备得更加充分。 6. AI已经成为资本开支增长的重要动力 Business capital expenditures—the seed corn of future economic growth—are rising rapidly. The four-quarter change in investment in equipment and intangibles has been around 9 percent, its highest growth rate since 2021. More than half of the cap-ex growth this year can likely be ascribed to the buildout related to Expectations for growth in both cap-ex and corporate earnings are running quite high. I will continue to watch the change in their growth rates, the second derivative. The follow-on effects on asset prices, business confidence, consumer income, and spending are equally important to gauge.市场对资本支出和企业盈利增长的预期都相当高。我将继续观察其增长速度的变化，也就是所谓的“二阶导数”。AI投资对资产价格、企业信心、消费者收入和消费支出所产生的后续影响，同样值得密切评估。 我的看法：Warsh这篇演讲把AI视为一种可能出现的新生产要素，这是一个偏积极的看法。美联储最关心的并非单纯的技术进步，而是AI对生产率、就业、产业收益分配、Token定价以及货币政策的长期影响。Kevin同时认为，AI投资已经对现实经济产生显著作用——今年美国资本支出增长的一半以上可能来自AI建设。