Curiosity - Research - Governance - Networks.

Joined March 2022
632 Photos and videos
to put things in perspective...
34
Amen.
I don’t care if “our enemies” get access to some super hacker AI. Just assume they have it and plan accordingly. I’m far more worried about AI being controlled by a small number of trillion dollar companies that decide who gets access and what you’re allowed to do.
1
26
God damn
Jun 9
JUST IN: Nvidia is now worth more than India
37
Uhthred Of Bebbenburg retweeted
you can just do things. you can just have a stupidly simple idea for a drone detection system, so obvious in your head that surely it already exists you can just spend the next weeks hyper focusing on research, only to realize it's not that simple, and maybe you're onto something you can just join a European defense hackathon and get the parts for your first prototype delivered 1 hour before it starts you can just spend 48 hours manically soldering, 3D modeling, printing, hot gluing, taping, cutting, coding firmware, fixing bugs, unrolling fiber optic cables, testing with a mini drone, building a pitch deck, and demoing it on stage where it kind of works but doesn't quite detect the drone you can just apply to an official FPV drone detection crash test in Ukraine, run by Ukraine's Ministry of Defense, without even having a working prototype yet, and get officially invited a few weeks later you can just realize you now have one month to build a working version from scratch, pack all your tools, and fly to the middle of Romania, to your cofounder's house, to spend 3 weeks hacking from 10am to 3am while frantically ordering more parts and bashing your heads against the infinite struggle of building hardware, electronics and software all tightly coupled together you can just try to open a company and a bank account in Portugal while in Romania, and fight the bureaucracy so hard that your bro has to call his mom to physically walk into a civil registry office in the countryside of Portugal, so she tells the registry lady to call you, so you can unlock the company formation process you can just fly back to Portugal to open the bank account in person, go to a Portuguese army innovation conference, and realize how hard it is to innovate in defense without being at war, and how far most European armies still are from the reality of what's happening in Ukraine right now you can just track down and buy the only Mac Studio M4 Max in Portugal, second hand, from a guy in the middle of the country, because it's sold out everywhere with an 8 week waiting list, since it's the only computer with enough compute and low enough power draw to live inside our portable field hub and detect FPV drones in real time you can just decide to travel to a country at war to test your product, and have the trip land just one week after the largest combined Shahed and cruise missile attack that country had seen in over four years of war you can just fly to Budapest, meet your cofounder and spend 2 days in a tiny hobbit house outside the city assembling and soldering the last sensor node, then drive last minute to a field full of wild horses to test the full system with a simple DJI drone, not really sure if it's going to work on the crash test you can just race to the Budapest train station straight off the field, buried in luggage, and board a 20 hour overnight train to Kyiv all by yourself, then talk your way past the Hungarian border police when they get scandalized that a Portuguese guy is rolling into wartime Ukraine with what looks like a weapon in a huge peli case, when it's really just a computer and homemade microphones connected by fiber optics you can just get to Kyiv and spend 2 days locked in your hotel room finishing the system hub, failing to get a GPS fix indoors to test the setup from your hotel bed, while air raid alerts go off every day and you head down to the shelter to wait them out with the other guests, passing around a plastic cup of Ukrainian champagne from Crimea you can just rent a car on the outskirts of Kyiv, get lost because the address was missing one letter after the number, walk 20 minutes at night to find the rental office, and finally get a reliable Skoda to drive to the crash test site at an undisclosed location you can just offer to pick up another participant from the train station on the morning of the test, whose train runs late because it had to be evacuated midway, and strike up a great conversation about defense, AI, Ukraine and acoustic detection the whole way there you can just show up late because you had to navigate to the site on pure vibes and old school map reading, and then start setting up alone while nothing seems to be working you can just beg the organizers for a stronger powerbank to feed the hub, realize after an hour of troubleshooting that the 4th node is dead (and you need at least 4 to detect anything), swap its GPS module right there in the field, and watch it all come alive 5 minutes before the first FPV test flight you can just spend the whole day in the field, eating dust under sun, wind and rain, alongside a field of other manufacturers all chasing the same problem in their own way, while your system hums along detecting and tracking real combat-grade FPV drones in real time, 50 meters out, plotting them on a map like radar, built from a pile of prototyping parts hacked together beautifully, to become a passive acoustic system that nothing can jam, made of 4 homemade microphones connected by fiber optics to a central compute hub you can just not have the official numbers yet, because the organizers are still processing them, but know exactly what you saw with your own eyes, a handmade system tracking combat drones in real time in the middle of a war and use that as fuel to drive you even more obsessed with cracking this problem you can just then take an overnight train to Lviv with all your gear, to compare notes with one of the leading Ukrainian acoustic-detection companies, only to evacuate the station over a bomb threat with your suspicious-looking giant peli hardware case you can just carry on toward Poland, meet a lovely Ukrainian soldier on the train who hands you drip coffee and gives you genuinely great product feedback, but ultimately fails to convince the border officer that you're not transporting a bomb you can just be pulled off the train with all your luggage and spend the next 12 hours questioned by 3 different teams of customs officers, in varying levels of broken english, about who you are, what you're doing, and what every single piece of electronics you're carrying actually does you can just have your whole prototype held at the border for further inspection, maybe to be returned in a couple of months, because nobody could quite believe it's just a regular computer and a bunch of homemade microphones connected by fiber optics you can just find a way to still make it to Vilnius in time for day 2 of the NATO-Ukraine innovators forum, despite losing your flight, and schmooze your way onto the stage to pitch in front of a panel of European defense VCs, running on coffee and zero sleep, in a way that makes sure they remember you you can just do all of that in 80 days, from the initial idea in your head to a working prototype, built with your own hands, that detects and tracks fiber optic guided FPV combat drones in real time, the ones that cannot be detected any other way, in a real test in the middle of Ukraine, so that fewer soldiers die at the frontline to this new class of weapon that has redefined modern warfare forever you can just start doing things and end up with a passive, electronic-warfare-immune acoustic detection system for FPV drones, cheap enough to blanket the frontline, holding up under live battlefield conditions in the most battle hardened country on earth right now you can just do things.
An FPV combat drone, detected and tracked in real time by sound alone. Four passive acoustic sensors connected by fiber, plotting it on a map like radar. Emitting zero RF, with nothing to jam. Silent Mesh is built to detect fiber-optic guided FPV drones, the deadliest new threat at the frontline, invisible to every other system. You can hear them before you see them.
17
18
84
10,813
Common sense ain't so common 🙄 !
A pause continues to be utter and complete nonsense and it always will be. 1. Let's make planes safer by not making planes! 2. What exactly happens in a "pause"? Do labs get to keep working and we all just get to sit on our hands waiting for this eureka moment? I guess someone gets to keep their checks while we just bring the economy to a crashing halt based on a few people's vague ideas about some imaginary future problems that they came up with while huffing glue and reading Dune! 3. Who will fund the labs when they are not putting out new products? I guess VCs will continue to just give them 100s of billions out of the goodness of their hearts! 4. What actually justifies a pause? Apparently, the wonderful world of imagination! Theoretical future problems that haven't happened yet. Like massive job losses! Um, jobs are increasing including in the areas most affected by AI, like coding, so I guess not that. So what? I know the jobs apocalypse is coming because I can imagine it and imagination is reality, right? Maybe advanced AI weapons? Ah, so the government will pause weapons research too? Well no, they will keep doing that anyway because they always do that. It's what governments do. Okay so we're going to ban Chat Bots while the government keeps making weapons? That should solve everything we're worried about with AI! Well then what about recursively improving models that grow to superintelligence overnight? Yeah that's not really a thing. That's the plot of an Avengers movie. Models are bound by the same real world constraints we are like compute (brains/chips) and time (will this drug have side effects in twenty years can only be known in twenty years) and fuzzy multiplicity (not right or wrong but right-ish and wrong-ish means you can't make a reward signal for "is this the right decision for my business") and the subject/object paradox (the thing improving is judging its own improvement. Yeah chew on that one for a bit.) But I imagined AI overcoming every real world constraint instantly so it's true! 5. How would we know we did everything we needed to do in a pause? How do we know when it's over? We dont. We just want a pause now because we want it! Don't you see my pause ⏸️ emoji? It's nice right! 6. Who gets to decide we are ready to unpause? The government or the people! Great, because vague ass platitudes like this always go well for concrete policy design. Looks, none of this is real. It's theater. It's not real policy. It has no basis in reality. It's a mass hallucination. It's pushed by people who believe in magic and magical solutions. And anything that comes out of it will do infinitely more damage than the imaginary thing they were trying to protect us from in the first place.
19
Uhthred Of Bebbenburg retweeted
Before the week ends, let's acknowledge one of the most INSANE week ever for open AI, with 25 notable open-weight drops across every modality: 🧠 LLMs → NVIDIA Nemotron 3 Ultra: 550B hybrid Mamba-MoE, only 55B active, 1M context, MMLU 89.1. NVFP4 variant claims ~5x throughput on Blackwell. First openly-weighted 550B hybrid Mamba-Transformer, closing the gap with frontier closed models. → Google Gemma 4 12B: fully open dense any-to-any (text/image/audio/video), 256k context, encoder-free, 140 languages, AIME 2026 at 77.5. Shipped with a 23-checkpoint QAT wave (mobile ONNX MLX). Most deployable model of the week. → StepFun Step-3.7-Flash: 198B sparse MoE VLM, ~11B active, SWE-Bench PRO 56.3. Apache 2.0. → Liquid AI LFM2.5-8B-A1B: edge MoE, just 1.5B active, 128k ctx, MATH500 88.8, MLX-ready. Best on-device option this week. → JetBrains Mellum2-12B-A2.5B-Thinking: their first open MoE, near-Qwen3-14B coding at 2.5B active. Apache 2.0. 🎨 Image gen (the surprise of the week) → Ideogram 4: their FIRST-EVER open weights. 9.3B flow-matching DiT trained from scratch. #2 overall behind GPT Image 2, top open-weight model on Design Arena LMArena. Strongest open checkpoint for text-rich images, full stop. It has taste. Still can't believe this is open weights. 🔊 Audio & Speech (a breakout week for open TTS, 4 labs shipped) → Boson Higgs Audio v3 4B: 102 languages, 21 emotions, singing/whispering/shouting, sub-second TTFA. → RedNote dots.tts: the only fully continuous (no codec) open TTS pipeline, Apache 2.0. → Google Magenta RealTime 2: real-time music gen, <200ms latency, text audio MIDI. multimodalart ported it to PyTorch within hours with live ZeroGPU demos. → NVIDIA Nemotron-3.5 ASR: 600M streaming, 17x more concurrent streams vs Parakeet RNNT 1.1B. 👁️ Vision & VLMs → PaddleOCR-VL-1.6: SOTA document parsing at 1B params, Apache 2.0. → Baidu NAVA: 6.3B joint audio-video gen, best-in-class A/V sync, Apache 2.0. 🎬 Video, 3D & World Models → NVIDIA Cosmos3-Super: 64B omnimodal world model coupling action trajectories with video audio gen, for Physical AI. → JD JoyAI-Echo: up to 5-min multi-shot text-to-video on LTX-2.3. → ByteDance Bernini-R VAST TripoSplat (single-image-to-3D Gaussian splats, MIT).
87
409
2,757
527,792
you can't be 50 years old in 10 months ♥️@ylecun
Replying to @Dan_Jeffries1
Did some exponential-pilled bros finally realize that real-world processes have irreducible time constants and that you can't run the real world faster than real time?
39
and bailout..

ALT Money Rain Money GIF

President Trump said he is considering taking a government stake in leading artificial intelligence companies. Industry leaders will soon gather at the White House to discuss the idea, the president said. wapo.st/4e28MbJ
14
if i ever was a fan !
> CBSE invited 19-year-old ethical hacker Nisarga Adhikary to help identify security gaps in its IT systems *insert i am proud to announce linkedin post*
1
209
I would rather let someone get a sim card in my name, or use a duplicate phone than install this bs surveillance app. No thank you @GoI_MeitY #wtf
1
56
I guess the meme hit hard
Replying to @Hedgeye
“If we don’t include them in the index they’ll just take the IPO right down the street to the NYSE”
1
5
512
Every single statement is defensible in a court of law but..
The broligarchs did it. The market is broken. It’s all built on trust. There will be none left. SPCX is a battering ram that shatters the illusion that the broad market is a stable instrument that can be trusted with your retirement. Goldman Sachs is selling science fiction. Not colloquially speaking—actual science fiction. The company has been propped up by government welfare since the start. The only viable business is Starlink. It earned about $4 billion in 2025 on $11 billion in revenue. Everything else is hot garbage. The AI business is a failure. Added to Starlink, it makes SpaceX wildly unprofitable, losing $5 billion total last year. The total revenue for the combined company was $18 billion. That makes the $1.75 trillion valuation about 100X sales—which is absurd on a scale never seen before. The rest of the valuation is based on science fiction that doesn’t even deserve the category “hard sci-fi”—which is reserved for stories with some plausibility. The story is “data centers in space” which is impossible in ANY relevant timeline. There are numerous scientific, financial, and physical roadblocks with no known solution at this time—all assuming infinite demand. Nevertheless, the company and bankers are booking it into near term revenue projections. It’s just a straight up lie. They project a **10,000%** increase in revenue for the AI business which is FAR behind the leaders, and has lost every bit of talent it ever had. Elon had to shill out Colossus just to pump the numbers for a quarter. Also, he didn’t NEED Colossus, which should tell. you something about the actual demand. XAI is a total failure. Pumping it as a revenue engine is FRAUD. The company is set up to make Elon Musk the unfireable forever-dictator, which means he will be free to do whatever insane shit he wants with ZERO accountability. If he were the most stable man in the world that would still be stone-cold crazy. Elon Musk himself is a drug-addicted, pathologically racist, malignant narcissist who is trying to redpill his large online cult into giving up empathy, and to believe in great replacement theory. He invaded the government to steal our data. He turned away all refugees except racist South African farmers. He shut down USAID. He spent two weeks screaming about a Black actress. He is a sick Nazi who wants to rack up a bigger kill count. To cap it off, it says right there in the IPO documents that Elon Musk’s actual goal is a COLONY ON MARS and that he won’t be compensated until there are a MILLION FUCKING PEOPLE there. How does a bank sign off on a company which has a stated goal of doing something that is not possible? If the core story of the business is a ludicrous sci-fi yarn, what does that tell you about it? This is the nightmarish alternate reality that these banks and markets are FORCING US TO BUY. Finally, they’re going to send TWO MORE of these hurricanes of fraud barreling into the indices—OpenAI and Anthropic. It’s a recipe for destroying the markets. All of it is madness. It’s greed and fraud with a new level of shamelessness. Perhaps, just maybe, the deep lasting pain this crime will cause will wake the rest of us up.
17
Uhthred Of Bebbenburg retweeted
22 yrs ago today, after a long zoning dispute with local officials that ruined his business, welder Marvin Heemeyer had enough & created the Killdozer. He destroyed the mayor’s house, the judge’s house, town hall, the police station, & the bank - while avoiding hurting civilians or their property. Happy Killdozer Day to those who celebrate 🎊
865
7,514
40,970
1,286,954
One of the few good things to come out of crypto ❤️
Today, almost 100k educators in Catalunya are deciding what happens next with the education strikes. And the vote is running on @vocdoni. This is precisely why we build open-source voting tech; this is how we democratize access to secure voting. 1/7 🧵
1
19
Goddamn..
Jun 4
Since arriving at its destination five years ago, our Perseverance Mars rover has collected data that hints at a history of past life on the Red Planet. Catch up on Percy’s biggest discoveries in this week’s episode of our Curious Universe podcast: go.nasa.gov/4x3NAuB

ALT A selfie image of the Perseverance Mars rover with the tan Martian landscape in the background. Credit: NASA/JPL-Caltech/MSSS

20
Macro snap ❤️
Monthly VC/LP debrief. What I actually saw in May 2026: 1/ SF is in full gold rush mode again, but history says the current winners won't stay on top forever. Every dominant technology eventually gets surpassed – newspapers, telecom, cable, Google in ads, IBM in computers. In AI the same pattern is already playing out: compute will hit walls, chips get dramatically more efficient, new energy sources emerge, and entirely new model architectures appear. The people feeling left behind today may just be early in a much longer cycle. (h/t @TurnerNovak) 2/ The largest $10B funds went from 140–150 collective early-stage deals per year in the SaaS era to 370–400 in the AI era. But the concentration is at the top of the market – top-decile rounds, known founders, proven operators. @kevinhartz calls it "option value": a small check today for the right to lead Series A tomorrow. The average seed round remains territory for EMs. 3/ We might be entering a Zombie VC era. ~85% of 2017–2018 vintage funds still haven't returned 1x DPI after 7–8 years. Median DPI sits at $0.34 on the dollar, while median IRR for the same cohort looks respectable at 11.6%. Paper returns hide the reality. The liquidity window opening over the next two years will be the moment of truth for most of these funds. 4/ @SpaceX IPO might be the single largest DPI event in VC history dropping into the lowest-distribution moment in venture capital history. @foundersfund alone, with an early $20M check in 2008, could return $60B (~3000x). When that capital hits LP accounts, it needs to be redeployed and that will circulate a new wave of fundraising for the same funds and fresh allocations from LPs who finally have liquidity to work with. 5/ The @cerebras IPO was the first real data point on crossover returns after two years of everyone writing off the model – both early-stage VCs and late-stage crossover funds made money on the same company, and LP conversations shifted from "do we have any exposure to the winners" to "how do we get into the next one." The same strategy that was declared dead in 2022-2023 got fully rehabilitated by a single exit. (h/t @MeghanKReynolds) 6/ Monte Carlo across 1,391 VC funds: concentrated portfolios (15 companies) and diversified ones (100 companies) produce the same average fund return – 2.44x. But compounded across multiple vintages, diversified wins: 2.25x vs. 1.78x. Concentrated funds carry more variance per fund, and variance drag compounds against you over time. The extreme outcomes (15x ) are almost exclusive to concentrated funds but the probability is tiny either way. (h/t Steve Kim) 7/ EM activity is showing the first real pulse in years. @cartainc logged 78 new US venture funds in the $10M–$100M range in Q1 2026 – a 34% jump from Q1 2025. Still well below the 2022 peak of 147, but the post-winter bottom might finally be in. The managers raising right now are doing it without a favorable macro, without easy LP recycling, and into a market where mega-funds are more active at seed than ever. (h/t @PeterJ_Walker) 8/ 76% of all EM-focused FoFs are American. The entire addressable market for a Fund I or Fund II isn't 132 FoFs – it's roughly 33. The other 100 exist, but Classic and Government-Led FoFs structurally can't anchor an early-stage vehicle: the check size doesn't justify the overhead, and a pension board can't be sold on a first-time manager without a track record. Geography and fund type filter out 75% of the market before the first meeting. (via @murphcapital) 9/ The 10-year fund is structurally mismatched with the assets mega-funds are holding. @SpaceX has been private for 18 years. @stripe for 15. For managers at that scale, @sequoia's move makes sense – open-ended, permanent capital, indefinite horizon. For small funds the logic runs the opposite way: the 10-year horizon enforced as a hard constraint, secondaries at Series C/D as the default exit, actual distributions on schedule. (h/t @credistick) 10/ There are only 3 positions that matter in a startup's cap table story: first investor, most helpful investor, biggest investor. Biggest is reserved for ~10 megafunds. First requires conviction most managers don't have – and LP preferences for concentrated portfolios often push against it structurally. So 90% of firms end up competing for "most helpful," which is why every pitch deck has a platform slide and every GP talks about their right to win oversubscribed rounds. (h/t @arian_ghashghai) Every month I track new fund launches, LP events, market reports, and what's actually moving in VC/LP. All of it in the @murphcapital newsletter: murphcapital.substack.com/p/…
1
60
🦔Microsoft's internal strategy document for its new AI assistant Scout says the explicit goal of phase one is to "make people addicted." The doc, obtained by 404 Media, outlines a three-phase plan from "addictive app to agentic platform." The tool sits on your desktop, manages your calendar, triages your inbox, files expenses, and acts on your behalf. It requires access to your accounts and files. Security and compliance are things to "figure out" later. Nadella already uses it. My Take After everything this week, I think this document accidentally explains the entire AI business model. Not just Microsoft's, everyone's. The product can't sustain itself on current pricing. We know that because Copilot just proved it on Monday. The unit economics don't work at flat rate. So the play is to get people locked in before the real bill arrives. Make the tool essential to how you work, let your company cut the people who used to do those tasks, and by the time consumption pricing kicks in, walking away costs more than paying up. IBM's CEO just told us the industry needs $6 to $8 trillion in capex to chase revenue he says doesn't exist. Google diluted shareholders to fund a buildout it can't cover from cash flow. Oracle fired 30,000 people during a record quarter to redirect salaries into data centers. And Microsoft's answer to all of that is an internal doc where step one is addiction. They're not selling the product on value. They're selling dependency. Get people hooked before anyone calculates what it costs to run, and make sure they can't leave once they find out. A product that needs addiction to survive is a product that can't survive on its own. Hedgie🤗 404media.co/microsoft-wants-…
28
Cap table due diligence ❤️
Replying to @futurenomics
Series B was led by SBF
1
1
290
Commodity is getting pricier, get in on the shop - NOW. Slickest hustler of my time 🙂
$NVDA CEO Jensen Huang says future AI data centers could cost $80B to $100B per gigawatt.
28