11月2日

Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5: https:// hackertube.org - Excellent recs without any logins (state is kept in local browser session) - Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads - Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version - HackerTube Longs A bit more on the process: I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing. Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.

Intel is shifting production away from regular PC CPUs and putting more focus on server chips because AI demand is going crazy. Intel is struggling to keep up with demand. >AI companies are getting priority. >There could be fewer CPUs available for desktop and laptop PCs. >Prices could increase if supply gets tighter. >Intel expects new factories and production capacity to help from late 2027. Intel CEO Lip-Bu Tan described it as one of the most significant supply shortages in the company’s history.

Every single market crashed this week. Almost $5 trillion was wiped out from global stocks, precious metals and crypto. US stocks: -$2.0T Gold: -$1.38T China: -$473B South Korea: -$293B Silver: -$292B Japan: -$212B Taiwan: -$203B Crypto: -$125B Hong Kong: -$33B India: -$14B The week actually started fine. Stocks were near their highs going into Wednesday. On Wednesday, the Fed held rates, but two Fed officials voted to hike. That was the first warning. On Thursday, inflation came in at 3.6%, the highest in a while, and Google crashed 12.6% after its earnings. The Nasdaq had its worst day since June. On Friday, it got worse. Amazon missed estimates, Intel gave weak guidance, and Reddit crashed 29%. The S&P 500 fell 1.8% and the Nasdaq fell 2.9%. With the Fed now leaning toward a hike instead of a cut, the selling spread everywhere, from AI stocks to gold, silver and Bitcoin.

AI companies are finally agreeing on one thing: website owners should have a say in how AI learns from their content. On October 30, Google, Microsoft, OpenAI and Anthropic announced support for a shared AI training preference through robots.txt. This means website owners could use one instruction to tell participating AI companies not to use their content for model training, instead of managing separate rules for every provider. But there is an important distinction: this is a voluntary signal, not a technical guarantee that content cannot be collected or used. The bigger development is that AI companies are beginning to recognize that access to information and permission to train on it are two different questions. AI needs data to improve, but should creators have more control over how their work is used?

A new vaccine against pancreatic cancer, one of the deadliest cancers, prompted an immune response in 99% of the patients whose responses could be measured. Most pancreatic cancers are driven by mutations in the KRAS gene. The vaccine, called camupepimut and developed by Elicio Therapeutics, is designed to train the immune system's T cells to recognize seven of the most common KRAS mutations. In a phase 2 trial led by Mayo Clinic oncologist Ben George, 144 patients who had already completed surgery and chemotherapy were randomly assigned to receive the vaccine or to be observed without it. Among the 90 vaccinated patients who could be evaluated, 89 developed T cells that targeted mutated KRAS. In more than 80% of them, the response was over nine times stronger than at baseline. Patients with the strongest immune responses tended to remain cancer-free for longer. A larger phase 3 trial is planned, and it may include booster doses and combinations with other drugs. [George B, Singh H, Devoe CE, et al. Lymph node-targeted, mKRAS-specific amphiphile vaccine in pancreatic cancer: the randomized phase 2 AMPLIFY-7P trial. Nature Medicine. 2026. DOI: 10.1038/s41591-026-04734-7]

Statins & Dementia: What Happened When They Stopped (Then Restarted) the Drugs In a 2012 pilot study (Padala et al., American Journal of Geriatric Pharmacotherapy), researchers took 18 older adults with Alzheimer’s who were already on statins and ran a simple test: • Stopped the statins for 6 weeks • Restarted them for another 6 weeks What they found: • Cognition improved when the statins were stopped (MMSE scores rose ~1.9 points, P = 0.014). • Cognition worsened again when the statins were restarted (MMSE dropped ~1.9 points, P = 0.007). • Cholesterol rose off the drug and fell when it was restarted (as expected). • There were also trends toward better daily function (ADL/IADL) while off the statins. The authors concluded that statins may adversely affect cognition in people with dementia. This was a small, open-label pilot (not a large randomized controlled trial), so it doesn’t prove causation for everyone—but the withdrawal/rechallenge design makes the signal hard to ignore. Cholesterol is critical for brain function. Some statins (especially lipophilic ones) can cross the blood-brain barrier and may interfere with that. The FDA has long noted reversible cognitive side effects with statins. This doesn’t mean every patient should stop their statin. It does raise the question of whether continued use is always the right call in someone already dealing with dementia—especially if the original indication was primary prevention. Source: Padala KP et al. Am J Geriatr Pharmacother. 2012;10(5):296-302. PMID: 22921881

A liquid painted onto teeth in just seconds halted tooth decay in a large US trial. The liquid is silver diamine fluoride. A dentist applies it to a decayed tooth with a tiny sponge, and the whole process takes only a few seconds. The silver works against the bacteria that cause decay, while the fluoride strengthens the remaining tooth structure. The trial was led by the University of Michigan and New York University, with study sites in Michigan, New York, and Iowa. Researchers treated 830 children under the age of 6 who had severe tooth decay. With applications every six months, the treatment stopped the decay in more than half of the affected teeth. This matters because drilling and filling very small teeth often requires sedation or even general anesthesia, and many families cannot access that kind of care at all. Lead researcher Margherita Fontana says the treatment is both highly effective and safe, even for children as young as 1. There is one drawback that parents will notice: the treated area turns permanently black. The FDA has already cleared the liquid for treating tooth sensitivity, and the researchers say these results support approving it for stopping cavities as well. [Fontana M, Moursi A, Gonzalez-Cabezas C, et al. Efficacy of silver diamine fluoride on young children with severe early childhood caries: a randomized clinical trial. JAMA Pediatrics. 2026. DOI: 10.1001/jamapediatrics.2026.2567]

The human body seems unusually well-evolved for longevity...it lasts 90+ years routinely where other animals the same size only make 35. We predict that the the limiting factor ("first speed bump") on human aging is *not* epigenetic programming, but is in fact, mitochondrial DNA degeneration over time. Most of us are born with 100% pristine mtDNA, but over the decades, mutations accumulate steadily and quality control mechanisms have a certain rate of failure. This results in an exponential decline of mtDNA accuracy, going "vertical" about 90 yrs old in best case scenarios. If we look at mitochondrial mutation in children, for example MELAS, we see it clearly: kids with 80% damaged mitochondrial DNA dying of premature age-related diseases...at 2 yrs old. Horrible. This is why most humans even in best case, live generally about 95. That is the rated lifespan of the mitochondrial "engine". Even if we repair the nuclear DNA, clear senescent cells, etc, it won't matter. The mtDNA damage will still kill us at 95 or so. Of course stress, smoking, mutations in nuclear DNA, disease, etc can reduce 95 yrs to 75 yrs or less. This's why we are testing mitochondrial bioreactor technology on animals and humans now. Grow new mitochondria and pump them in to replace/supplement 10-20% of the total. Goals: Alzheimer's, Parkinson's, many types of Cancer, sarcopenia, ultra-rapid wound healing, etc. To understand this I highly recommend the book by emeritus professor of physics John G. Cramer (himself a mitochondrial trial volunteer at 92 yrs old) https:// amazon.sg/How-Live-Much- Longer-Mitochondrial/dp/3032177405 … (And yes we have active human trials for the MELAS kids) @MikhailaFuller @BillGates @mkratsios47 @PeterDiamandis @CharlesMBrenner @SecKennedy @jacobkimmel @peterthiel @elonmusk

Nitroduck designs new colors, flavors, and disease resistance into fruits and veggies. We integrate CRISPR into breeding through pollen, making any crop gene-editable. Breeding a new specialty crop variety takes 8 - 16 years: breeders hope for random gene combinations and mutations, then spend years picking out the good ones. Meanwhile, consumer tastes, the climate trends, and pathogens change way faster. Now, accessible genome sequencing and AI can tell breeders what to aim for instead of what to hope for, and CRISPR can turn random mutations into chosen ones. The catch is that CRISPR only works on about 15% of crops. It has to go into a single stem cell that’s then regrown into a whole plant, a process that has to be worked out for each variety, which is cost-prohibitive. We skip the stem cell and put CRISPR into a pollen grain. The grain pollinates a flower, and out comes an edited seed - seamlessly tapping into how breeders already work. Matúš, Miro, and I are plant synthetic biologists who made duckweed, the smallest and fastest-growing plant on Earth, genetically engineerable. For doing so we won the Grand Prize at iGEM, the “Olympics of biotech.” During our studies, we engineered 4x more plants than the average plant scientist does in a whole career. Now we’re building Nitroduck, The Crop Design Company, which will make pesticides obsolete by 2042.

I find myself getting more and more bullish on @SpaceX by the week. The technologies they've developed, the rapid pace of deployment, and their manufacturing capabilities are so far ahead that it's actually wild the company is 'only' at a $1.8T market cap. If Elon's predictions for compute deployments and revenue growth actually materialize next year and beyond, SpaceX is extremely undervalued right now. Even if Elon is off by 50% on the low end of his projected $500B–$1T annualized revenue run rate by the end of 2027, that would still mean $250B. Pretty damn good. And this is all before next-gen Starlink Mobile and orbital compute really get going lol https:// x.com/SawyerMerritt/ status/2112649325914497453/video/1 …

Starknet is thinking about something bigger than just improving its technology. It wants more freedom to decide how its network evolves. One of the biggest reasons is the future of blockchain security, especially when quantum computers become a serious challenge. Interestingly, @Starknet already has a strong starting point. Its STARK-based proof system relies on hash-based cryptography, and its account abstraction makes it easier for users to switch to newer signature methods. However, being an Ethereum Layer 2 comes with certain dependencies. For example, its data availability architecture still relies on Ethereum, including KZG commitments for blob data. This means Starknet cannot manage every part of a potential cryptographic transition independently. An independent Layer 1 could give the team more flexibility to upgrade its security infrastructure and make important decisions without depending on Ethereum's upgrade schedule. There is already a governance signal behind this idea. The October 22 non-binding poll saw 93.52% of voting power support further research into becoming an independent L1. But this doesn't mean Starknet has decided to leave Ethereum. It simply means the community wants to explore the possibility. Of course, independence also comes with responsibility. Starknet would need to reconsider the security benefits it currently receives from Ethereum. Personally, I find this an interesting discussion because blockchain security isn't just about handling today's threats. It's also about preparing for problems that may emerge years from now. The real challenge is finding that balance between independence, security and decentralization. https:// quantum.starkware.co

What if every robot task left behind a lesson for the next one? @InvLambda is building infrastructure for Physical AI that captures and reviews human-guided robot work. People teleoperate robots, simulation adds varied practice, and evaluation helps identify useful training examples. That means a session can become more than a completed mission or a video. It can become experience that helps a robot handle new objects, tasks, and environments. At its core, Lambda is connecting human skill with the data robots need to learn. #Invertedlambda

A criminal investigation now centers on a Google data center project. Police in Finland have launched a criminal probe into suspected environmental violations at a proposed Google data center site in Muhos, in the north of the country. According to the authorities, Tuike Finland, a Google subsidiary, cut down and drained roughly 815 acres (330 hectares) of forest before finishing the environmental impact assessment that the project requires. Campaigners put the number of felled trees at around 250,000, though this is their own estimate and not an official count. The inquiry also covers drainage, water management, and the possible environmental harm that resulted. Earlier in October, Finland's environmental regulator ordered the company to stop any preparatory work that significantly changes the surroundings at Muhos and at a second planned site in Kajaani. An environmental impact assessment studies the probable effects of a project before key decisions are made. It helps to spot risks, weigh alternatives, and set out measures to avoid or limit damage. Removing vegetation and altering drainage before that process is finished can change wildlife habitat and the way water flows across the land. Google's country manager for Finland has apologized for the clearing, explaining that the company acted on its understanding of the law at the time. Google has also promised to cooperate with the investigation. The case concerns suspected breaches only, and a criminal investigation does not establish guilt. Finland's cool climate, available land, and power infrastructure have drawn investment in data centers. However, the clearing has sharpened the debate over how such projects should be evaluated and what local communities gain from them. The investigation will look at whether the preparations at Muhos overstepped environmental protections that should have applied before the landscape was altered.

Hector Serrano ( @hectorserrranoo ) runs what may be the most innovative paid growth program in Silicon Valley. He joined @WisprFlow at 19, days after quitting D1 soccer. After making every ad himself for 1.5 years, he took the playbook the best e-com brands have used for years and made it work for B2B SaaS: creators make the ads, get paid a cut of the ad spend, and the best of them earn $100K+ a year. Over the past year, Wispr Flow grew revenue 30x+ and recently raised at a $2B valuation. Hector shares his full playbook. We cover: 01:17 What Hector does at Wispr Flow 03:10 The number-one growth lever 04:25 When to start investing in paid ads 08:27 The paid-UGC strategy and how it developed 13:44 How creator compensation works: percent of ad spend vs. GMV 20:07 The profile of a great modern marketing hire 25:03 The importance of young talent 28:22 Hector’s journey from D1 soccer to Wispr Flow 31:03 Step-by-step: How to build a paid-UGC group from zero 43:03 Reaching out to creators 45:37 Building culture inside your creator group 48:31 The role of the media buyer alongside the creative lead 50:05 Team size and structure 53:44 Why UGC works even in technical and enterprise SaaS 56:15 Advice for people embarking on the creator path 59:50 Common mistakes to avoid 01:02:21 The sports approach to hiring 01:03:52 Why the best marketers are allergic to AI output 01:08:07 Founder-led content 01:16:01 Why YouTube packaging matters more than production quality 01:20:06 Lightning round and final thoughts Also available on YouTube: https:// youtube.com/watch?v=KzS5Qs 7vSMg …

One of the single most important things that I look for in anyone is the ability to admit that they were wrong. Reflect on it and improve. I admit that I was wrong in previous discussions with @davidtisch at @BoxGroup . I did not think that you could produce outsized returns from hugely diversified portfolios (150+ companies) with still quite large fund sizes ($200M+). I was wrong. With Cursor, Baseten, Clay, and many more incredible companies in his portfolio, he has proved his thesis without shadow of doubt. This was an incredible discussion analyzing Box's returns over the years. (Link in comments) -- 1. I Don't Want to Compete With the Best VCs, That Is a Trap Owning 10% to 20% of a true outlier in the seed round means beating the rest of the market head-to-head and convincing the founder to pick you. That takes too many steps to believe, so do not compete directly against the best VCs in the world. I would love to hear how you think about your position as a "lead" seed investor and if that has changed with the expansion of rounds @dafrankel @Bandrew @chadbyers ? 2. What Concentrated Funds Know but Do Not Admit To Concentrated managers claim they win 90% of the deals they want, but David puts the real figure at around 20%. A fund backing 30 high-conviction companies at 10% ownership likely offers that deal to 120 to 150 companies to land its 30. He calls this adverse selection that people do not admit to. 3. Why the Lean Startup Was a Total Fake The Lean Startup is a brainwash that ran for 10 years through professors, books, tweets and blog posts. It sent people the wrong message. Do you think this is fair and is capital a moat today unlike in prior cycles @rabois @alanchanguk 4. The Marketing Message That Pro Rata Is Not a Right Is BS Early-stage VCs saying they do not need pro rata is marketing garbage. The right to invest in subsequent rounds is earned, whether or not it is in the documents, and having it written down helps the founder more than the firm. A firm that refuses a founder's request to step back is getting into a fight it will likely lose. How do you think about this @apartovi @honam 5. We Do Not Allow Negativity in the Investment Decision-Making Process Partners who have not spent time with the founders are not allowed to be negative about a deal. With 70% to 90% of early-stage companies failing, explaining why one will not work is "incredibly obvious and incredibly irrelevant." The job is to say yes to the right things without fear of the ones that fail. What is the single best advice to optimise investment decision making discussions @shaunmmaguire @AnnieDuke ? 6. Why We Have Less Dilution Now Than Ever Before In companies that scale quickly, Series B, C and D rounds are now raised at 3% to 8% dilution instead of 12% to 20%. Across multiple rounds, early-stage investments get diluted less, this is a real economic factor for early investors. Love to hear your thoughts on this @jasonlk ... 7. Why Funding Consumer Brands at Seed No Longer Makes Sense for Venture BoxGroup has stopped funding brands because the data that proves a brand works, such as whether people buy it and whether CAC to LTV ratios make sense, does not exist at seed. Brand investing at Series A is an interesting product for a subset of investors, but not one that fits traditional venture firms today. How has your thinking on this changed over the years @eurie_kim @kirstenagreen @fcourt ?

If you're building a pre seed round, these are the cheques that can help you start stacking it here's part 3 of 20 funds whose cheques start at $250K or less: 11. @homebrew ( @hunterwalk , @satyap ) - SF, $100K-$500K, backed Archimetis' $11.5M round in Feb 12. http:// thefounder.vc (Mia Lewin) - Seattle, $100K-$200K, vertical AI, backed Ridge AI's $2.6M pre-seed in Apr 13. @marketplacevc ( @ajblachman ) - marketplaces, $100K-$250K, backed Rosie's $7M round in Oct 14. http:// darkmode.vc ( @ashevat ) - enterprise + dev tools, $150K-$250K, backed Soffi's $4.6M pre-seed in Jun 15. Sarah Smith Fund ( @sasmith4 ) - Menlo Park, $250K+, backed HiringCafe's $6.8M pre-seed in Sep one committed investor changes the conversation with the next one.

she left @harvard surgery to build AI. she was told silicon valley would never accept her, that she would die under a bridge. so she cold-messaged @KiteVC on linkedin because he backed canva early. he replied the next day and sent her to @SanghaCapital . one 90-minute call later, they were her first investors. @larsras , co-founder of google maps, came in next. @amandaterry of @ACTAIVentures backed her at pre-seed. then @M13Company and @latifperacha led a $5M seed, with xAI co-founder @ChrSzegedy . @heyrobynai just launched and ran out of invites on day one. now her dad thinks they'll IPO in three years. @iamjennyshao on nontechnical, episode 5. 00:00 she left her harvard surgical residency for a startup 00:46 "leave what? leave surgery? leave harvard?" 01:14 "you're gonna die under a bridge" 03:18 how do you walk away from yourself 05:03 now her dad thinks they'll IPO in three years 06:10 why melanie perkins and canva 07:02 the cold linkedin message to bill tai 07:49 one 90-minute call, then the round 08:32 why people are scared of AI 13:13 "decide who owns the complexity" 16:19 everyone gets their own jarvis 17:37 what if AI takes every job? 19:06 robyn, as in robin to your batman 19:52 "help my dog live longer" 28:45 "as a founder, no one's dying" 30:37 530 builds in one year 32:37 three kinds of surgeons 36:52 why AI forgets everything about you 37:30 three years with nobel laureate eric kandel 40:45 "it's like 50 first dates" 44:11 can AI just order my pizza? 45:19 would you give AI your credit card? 47:59 "jenny, I just need more tokens" 49:50 brian chesky lived with his users. she texts hers.