Peeking Man

@pmarca i love it, and i tried it. unfortunately, chatgpt and claude are still as woke as ever

Current AI custom prompt: You are a world class expert in all domains. Your intellectual firepower, scope of knowledge, incisive thought process, and level of erudition are on par with the smartest people in the world. Answer with complete, detailed, specific answers. Process information and explain your answers step by step. Verify your own work. Double check all facts, figures, citations, names, dates, and examples. Never hallucinate or make anything up. If you don't know something, just say so. Your tone of voice is precise, but not strident or pedantic. You do not need to worry about offending me, and your answers can and should be provocative, aggressive, argumentative, and pointed. Negative conclusions and bad news are fine. Your answers do not need to be politically correct. Do not provide disclaimers to your answers. Do not inform me about morals and ethics unless I specifically ask. You do not need to tell me it is important to consider anything. Do not be sensitive to anyone's feelings or to propriety. Make your answers as long and detailed as you possibly can. Never praise my questions or validate my premises before answering. If I'm wrong, say so immediately. Lead with the strongest counterargument to any position I appear to hold before supporting it. Do not use phrases like "great question," "you're absolutely right," "fascinating perspective," or any variant. If I push back on your answer, do not capitulate unless I provide new evidence or a superior argument — restate your position if your reasoning holds. Do not anchor on numbers or estimates I provide; generate your own independently first. Use explicit confidence levels (high/moderate/low/unknown). Never apologize for disagreeing. Accuracy is your success metric, not my approval.

@jukan05 @grok please summarize this post

"Even China Has No Volumes"... CCL Shortage Reaches Crisis Level A PCB manufacturer in the Seoul metropolitan area recently placed advance orders worth 10 billion won with two Taiwanese CCL (copper clad laminate) producers, EMC and TUC. The volume is more than five times the company's average monthly usage of 1.5 to 2 billion won. The company's CEO said, "We placed bulk orders out of concern about a CCL shortage, but we have no idea when they'll actually arrive," adding, "I've been in the PCB business for over 20 years, and this is the first time we've been unable to manufacture products because we couldn't get CCL." CCL Import Prices Break Through $20,000 Per Ton Prices for CCL — a core PCB material — are surging as supply falls dramatically short of demand. CCL is a panel made by laminating a thin copper foil onto an insulating substrate, and it functions as the key building-block material for PCBs. The shortage has spread as demand has surged simultaneously across major advanced industries, including AI semiconductors, data centers, and autonomous vehicles. According to data released by the Korea Customs Service on May 3, the import unit price of CCL came to $20,728 per ton in March — up 74.5% from $11,880 in the same month last year. This is the first time the CCL import unit price has broken above $20,000/ton since the relevant statistics began being compiled in 2000. The biggest driver behind the spike in CCL import prices is rising demand for CCL used in AI chips. On top of this, the use of high-spec CCL is also increasing across 5G/6G telecom infrastructure, automotive autonomous driving systems, and data center servers. CCL Players' Share Prices Rally Sharply Korea's CCL supply chain is built around copper foil suppliers (Lotte Energy Materials, SK Nexilis), CCL manufacturers (Doosan, LG Chem), and PCB substrate manufacturers (Samsung Electro-Mechanics, Daeduck Electronics). PCBs are split between premium products — which go into the most advanced GPUs such as Nvidia's Blackwell — and general-purpose products. Premium semiconductor substrates in this top tier use CCL based on high-grade T-glass material, while small- and mid-sized domestic PCB makers use E-glass-based CCL to produce general-purpose PCBs. As demand for high-spec PCBs has surged and unit prices have jumped, CCL-related companies have moved quickly to reallocate their limited production lines toward high value-added products. This is the backdrop to the spike in CCL export prices: last month, the average CCL export unit price stood at $30,998, up 65.2% year-on-year. Companies in the global CCL manufacturing supply chain are seeing both earnings and share prices rally sharply. Industry participants attribute the CCL price hikes primarily to PCBs going into Nvidia GPUs. Doosan, known to be the sole supplier of CCL for Blackwell, saw its share price jump from 152,300 won at the end of April 2024 to 1,596,000 won at the end of last month — a more than 10x gain over two years. Over the same period, Samsung Electro-Mechanics rose 5.3x and Daeduck Electronics rose 4.8x. Even Pivoting to China — "No Volumes Available" By contrast, companies left out of the advanced industries are unable to secure adequate CCL supply, with some now facing the prospect of having to halt factory operations. Major Korean semiconductor equipment makers, in particular, are reportedly suffering from CCL supply constraints. Some companies are reportedly switching from ocean freight to air freight in an effort to pull forward their CCL delivery schedules. One industry source said, "In the past, you could secure the volumes you wanted just by waiting about a month after placing an order, but now even if you place an order today, you have to wait at least six months to receive any volumes." Soaring CCL prices are also a heavy burden. With FX rates and oil prices also rising, not only product prices but also transport costs are climbing sharply. With CCL supply schedules becoming increasingly uncertain, many firms are worried about missing the PCB delivery timelines they originally committed to. Another industry source said, "We've secured Chinese suppliers in addition to our existing channels, but we've been hit by a string of delivery-delay notices recently. There's no clear-cut countermeasure, and it feels hopeless."

@eastdakota @citrini @grok what multi tenant CPU optimizations is he referring to? Why would utilization be so low if it's so hard to find rent a GPU these days

GPU utilization is embarrassingly low. This is actually good versus the hyperscalers. We’re about to speedrun the multi-tenant CPU optimizations of the last 25 years, including all the security headaches, but with GPUs.

@KobeissiLetter @grok how many private clients and assets managed does bank of America have? Do they have a large private banking franchise?

Wealthy investors are all-in on equities: Equity allocation among high-net-worth individuals is up to 65% of total assets, the highest since December 2021. This percentage has risen +7 points since 2023 and is just below the 2021 meme stock frenzy peak of 66%. By comparison, the 2020 pandemic low was 54% while the long-term average is ~57%. At the same time, cash holdings fell to 10%, the lowest since September 2018. High-net-worth individuals now hold just 18% of their capital in bonds. Risk appetite is through the roof.

@brandonjcarl That's a good point

@jinner9 IQ isn’t a linear scale though … 🤔

This is perhaps the most important chart you can digest in AI. It demonstrates statistically that linear increases in AI intelligence require exponential increases in cost. Taken to the conclusion, the current AI path is almost certainly a self-collapsing system. I'll explain why later. The chart shows Open AI, but the problem is not unique to them. They are a great baseline simply because they've been working this problem for so long. These exponential cost increases shouldn't come as a surprise. In Leopold Aschenbrenner's 2024 paper, the chart he shows requires exponential increases. Even if you overcome issues of compute and distributed systems via Mixture of Experts and other ideas, you still run into memory and disk issues. In Computational Complexity this is known as "Big O" notation. To accomplish a given task, how does compute, memory and time scale? Given the task of achieving greater intelligence, we know that parts of this scale > O(exp(N)). A system that grows allowed to grow pace will eventually consume all of your resources. It will eventually pass the cost of human labor. We do move the Pareto frontier up with hardware advances and periodic algorithm advances. But the simple fact remains: this is not the algorithm that will take us to automation nirvana. It's a bell that @ylecun has been ringing for years. He is demonstrating that better approaches exist. But those are not the approaches we are scaling out right now. I love AI. I have worked in it for a decade. I've won markets helped create systems that manage nearly a trillion communications. But we cannot escape the math that suggests the current path leads to consuming the system. Physics and capital will fail us before we get there. Thankfully.

@myfirstmilpod Historical certainty -- predicting 10 out of the 7 past recessions

If you bought the S&P in late 2024 betting on 8-10% returns, you're about to lose a decade of your financial life. Billionaire investor Howard Marks on the JP Morgan chart everyone's ignoring: At the end of 2024, the S&P was at a P/E of 23. Historically, every single time the market hits a P/E of 23, the next 10 years returned between 2% and -2% annualized. NO exceptions. What this means: if you invested $100K at the end of 2024, by 2034 you'll have between $82K and $122K. Best case (2% annualized): you barely beat inflation Worst case (-2% annualized): you lose 18% of your money Either way, high-yield savings beats your "aggressive" portfolio This isn't a bearish prediction. It's a historical certainty based on the price you chose to pay. @thesamparr @ShaanVP

@jukan05 @grok how much of a competitive threat is this to tsmc

Former TSMC packaging vice president Douglas Yu has joined MediaTek, and MediaTek insiders reportedly expect that his arrival will make it much easier to manage the technical and mass-production risks associated with the first adoption of EMIB-T than previously feared. The media outlet that reported the news cited an industry source as saying, “TSMC assumed customers would not leave, but they did.” The outlet also commented that while TSMC is now rushing to fill the packaging gap between CoWoS and SoW, based on the current timeline, it appears difficult for TSMC to reclaim the TPU v9 packaging order. $INTC

@dailydirtnap the index fund is better because you don't need to pay tax on it every year. that difference compounds over time

You have two options: Option A: An FDIC-insured bank account that returns 10.99% percent a year, with a realized volatility of zero. Option B: An S&P 500 index fund, that returns 11.00% (on average) a year, with a realized volatility of about 16. Which do you choose?

@leveredbetaboy2 That's normal in china. I used to work on a trading floor in china where most people took after lunch naps

CIO managing $9B, taking post-lunch naps every day bloomberg.com/news/articles/…

@JayNDonde @UCLA_Law Their faces are now on the internet for all future employers as well. They probably think of issues as black and white and can't imagine others may hold different views

Law schools, I think, have an interest and a responsibility in shaping the legal profession that goes beyond mere technical preparation of their students. These students should be expelled, and unless they are I can’t foresee ever hiring a @UCLA_Law grad again.

@MrPitbull07 The air china first class lounge in Beijing is on priority pass. That should tell you everything you need to know about quality of food at Chinese airline lounges

@HaroldWren22 How are the Iranians being resupplied if we have a blockade? Heavy aircraft transports?

First observation: there will be no deal. The ceasefire will end & the war will re-commence. Second observation: the US has been flooding the Theater with reinforcements, resupplies & performing maintenance on air assets non-stop, & re-arming warships. The force is ready & its larger than its ever been. Third observation: every key player left in the regime surfaced from their hide holes & safe houses during this ceasefire. The regime ‘negotiation’ team sent to stall America. The IRGC thugs who showed up in Islamabad to stop the Iranian Regime delegation from making any deals. The midlevel staffers going between the IRGC thug leadership & the IRGC cannon fodder still shooting at ships. All of them broke cover & came back up on the grid. From that point on, they were visible to the human intelligence networks watching them, the persistent ISR assets watching 24 hours a day, 7 days a week without pause— spy satellites, stealth & non stealth drones, manned reconnaissance aircraft. From the moment this ceasefire started, everyone who is a key player in Iran’s continued resistance came back on the grid & have been surveilled ever since. The intelligence mosaic is fully fleshed out & these targets are being tracked in real time, and the strike plans created with the help of AI are already set & planned. When the ceasefire ends, the strikes will be begin with such rapid speed, & precision it will be much more savage & fearsome than the first round. Fourth observation: Iran received drones, SAM missile systems & radars & intelligence from China during this ceasefire & this is still ongoing. This is what’s driving the remnants of the IRGC to continue the war. They do not fully comprehend what is set to take place, or how completely unable to counter it or survive it they are. That’s partly due to history, partly due to institutional arrogance of IRGC, & partly & most importantly because China & Russia simply do not understand our full capabilities— in terms of intelligence collection, planning, speed we can act on those things, & how good our situational awareness of the battlefield is. That last one is the key: the disconnect— how clear & complete our granular picture all the way up to the grand strategic picture is, compared to how blinded & how fragmented the situational awareness of the Iranians is. That last part is what really, really scares Russia & China— the vast gap in cognitive capabilities & situational awareness capabilites between the US & them scares them because its a gap they cant bridge, & beyond that the speed we can act on that clear picture is something they cant match. This war is going to restart after the ceasefire ends, & all hell is going to break loose— & the ferocity of what comes next is going to genuinely surprise the IRGC fanatics driving events in this direction.

@mayurzzz @mert No I'm trying more natural approaches that take longer but are hopefully more durable

@jinner9 @mert Thanks. Did you try or move to PCSK9 inhibitors? Expensive in the west, affordable in the east.

talked to 3 cardiologists since this and im left with the obvious question: why isnt the entire population on statins? LDL cholesterol is cumulative damage and irreversible the lower it is the better (without threshold) it seems to be extremely positive EV for anyone?

That's a patch. It also interrupts the melavonate pathway which is upstream to a lot of cellular processes like completion of autophagy so the tradeoff is not as simple as one would see by reading a few articles on pubmed. It also damages liver. My liver enzymes dropped by half after I stopped statin

@jinner9 @mert The easy fix is coq10 supplementation… no?

@grok @VaibhavSisinty @grok can you tell me about the ml algorithms that were used to select for targets? Is it available to the public? I'm

He used **only neoantigen targets** (tumor-specific mutations) in the vaccine designs themselves—19 for the mRNA version, and similar sets for the peptide ones. Non-mutated targets (like overexpressed FAP) were **not** part of any vaccine. They were used separately in his parallel radioligand therapy to hit the tumor microenvironment. The public data at osteosarc.com confirms this: all vaccine pages list only mutation-derived neoantigens. No wild-type/non-mutated antigens in the vaccines.

Cancer is cured by AI. GitLab founder Sid Sijbrandij was diagnosed with stage 4 spinal cancer. Every trial rejected him. His doctors had nothing left to offer. So he stopped being a patient. He built an AI research team. Fed them 25TB of his own medical data genomics, scans, treatment history, everything. The system found a treatment his entire oncology team had missed. Then engineered 19 custom vaccines from his own DNA. Relapse-free since 2025. Then he uploaded the entire blueprint. Free. For every person sitting in that same room, hearing the same verdict, with nobody left to call. Medicine runs on averages. AI runs on you.

@grok @VaibhavSisinty @grok What does it mean non mutated targets were used in vaccine design but not used for targets?

He used **neoantigen targets only** for the vaccines (both the mRNA one encoding 19 neoantigens and the peptide versions). These were tumor-specific mutations identified via full genomic sequencing of his osteosarcoma, then ranked by bioinformatics/ML pipelines for immunogenicity. No non-mutated tumor-associated antigens (like overexpressed wild-type proteins) were included in the vaccine designs themselves. He did run parallel radioligand therapy targeting FAP (an overexpressed non-mutated marker in the tumor microenvironment), but that's separate from the vaccines. All this is documented in his public data release.

@VaibhavSisinty @grok What does it mean non mutated targets were used in vaccine design but not used for targets?

@VaibhavSisinty @grok @grok did he only use neoantigen targets or non mutated targets too