Sabitlenmiş Tweet
Iric
2.8K posts
Iric
@Iric_Chen
learning to learn openly. ceo, @FigoHealth, software-enabled renal care provider. partner, @Vida_Capital
🇳🇦·🇬🇧·🇹🇼 Katılım Aralık 2010
1.1K Takip Edilen342 Takipçiler
Iric retweetledi
The first step of any meaningful pursuit is to severely underestimate its difficulty.
English
Iric retweetledi
This isn't science fiction. This is real sequencing biotechnology just introduced by Roche. Wild and wonderful.
English
Iric retweetledi
@MariusCRoodt @WoganMay @ashtonshudson I forgot about this whole saga. Reliving it through Claude's summary is 👌🏽👌🏽👌🏽
English
I’m scared to shit post 🫣.
Given the way things are going, either people will believe it’s real (and attack me ); or worse, things are so crazy it might actually come true.
What a time do be alive.
English
Iric retweetledi
trying to explain OpenAI’s corporate structure after Elon put a $98B bid on the non-profit arm of the company…which is supposed to be worth 25% of the entire business…which is trying to turn into a for-profit and was recently valued at $300B after Masa put in $40B…which includes $15B he’s putting into the $500B Stargate AI infrastructure project (a separate entity) with Oracle and Softbank…which may not actually have all those funds…and Microsoft apparently still owns 49% of everything for $13B in Azure credits it committed years ago…but all of these deals will change based on a legal definition of “AGI”, which itself is coming in a few thousand days
English
We need a supercut of @chamath's evolution of views on the US government budget deficit and its increasingly unsustainable debt burdens.
When it was first brought up on the pod, he ridiculed @friedberg as Chicken Little for sounding the alarm bells.
Sven@SvenSchnieders
Chamath is a joke. Now that it’s safe to question the vaccine, he’s posting memes—but in 2021, he was imploring the gov to *mandate* it, sneering at skeptics “who frankly aren't doing it for medical or religious reasons; they're just watching Fox news and spouting off.” Supercut
English
Iric retweetledi
He is correct... and one of the most egregious examples of crony capitalism is the corrupt alliance between the Chinese-owned Smithfield Foods and senators like @SenJoniErnst , who are trying to push through a backhanded deal to FORCE gestation-crated pork back onto states which have already voted to ban it.
Pay attention to the "Farm Bill" they're trying to sneak in through Brooke Rollins - it's a complete f-you to Republican values of states' rights of self determination, it hurts small farmers, and it is terrible for American health.
Vivek Ramaswamy@VivekGRamaswamy
The biggest threat to capitalism is crony capitalism.
English
Iric retweetledi
Iric retweetledi
everyone is focused on running locally. very bullish for strava
English
Iric retweetledi
tried doing therapy with DeepSeek (same exact prompts I use with Claude) and it just told me to study harder and practice my violin and that I will feel happier once I finally graduate from medical school ???
English
Iric retweetledi
"Move 37" is the word-of-day - it's when an AI, trained via the trial-and-error process of reinforcement learning, discovers actions that are new, surprising, and secretly brilliant even to expert humans. It is a magical, just slightly unnerving, emergent phenomenon only achievable by large-scale reinforcement learning. You can't get there by expert imitation. It's when AlphaGo played move 37 in Game 2 against Lee Sedol, a weird move that was estimated to only have 1 in 10,000 chance to be played by a human, but one that was creative and brilliant in retrospect, leading to a win in that game.
We've seen Move 37 in a closed, game-like environment like Go, but with the latest crop of "thinking" LLM models (e.g. OpenAI-o1, DeepSeek-R1, Gemini 2.0 Flash Thinking), we are seeing the first very early glimmers of things like it in open world domains. The models discover, in the process of trying to solve many diverse math/code/etc. problems, strategies that resemble the internal monologue of humans, which are very hard (/impossible) to directly program into the models. I call these "cognitive strategies" - things like approaching a problem from different angles, trying out different ideas, finding analogies, backtracking, re-examining, etc. Weird as it sounds, it's plausible that LLMs can discover better ways of thinking, of solving problems, of connecting ideas across disciplines, and do so in a way we will find surprising, puzzling, but creative and brilliant in retrospect. It could get plenty weirder too - it's plausible (even likely, if it's done well) that the optimization invents its own language that is inscrutable to us, but that is more efficient or effective at problem solving. The weirdness of reinforcement learning is in principle unbounded.
I don't think we've seen equivalents of Move 37 yet. I don't know what it will look like. I think we're still quite early and that there is a lot of work ahead, both engineering and research. But the technology feels on track to find them.
youtube.com/watch?v=HT-UZk…
YouTube
English
Iric retweetledi
Iric retweetledi
I don't have too too much to add on top of this earlier post on V3 and I think it applies to R1 too (which is the more recent, thinking equivalent).
I will say that Deep Learning has a legendary ravenous appetite for compute, like no other algorithm that has ever been developed in AI. You may not always be utilizing it fully but I would never bet against compute as the upper bound for achievable intelligence in the long run. Not just for an individual final training run, but also for the entire innovation / experimentation engine that silently underlies all the algorithmic innovations.
Data has historically been seen as a separate category from compute, but even data is downstream of compute to a large extent - you can spend compute to create data. Tons of it. You've heard this called synthetic data generation, but less obviously, there is a very deep connection (equivalence even) between "synthetic data generation" and "reinforcement learning". In the trial-and-error learning process in RL, the "trial" is model generating (synthetic) data, which it then learns from based on the "error" (/reward). Conversely, when you generate synthetic data and then rank or filter it in any way, your filter is straight up equivalent to a 0-1 advantage function - congrats you're doing crappy RL.
Last thought. Not sure if this is obvious. There are two major types of learning, in both children and in deep learning. There is 1) imitation learning (watch and repeat, i.e. pretraining, supervised finetuning), and 2) trial-and-error learning (reinforcement learning). My favorite simple example is AlphaGo - 1) is learning by imitating expert players, 2) is reinforcement learning to win the game. Almost every single shocking result of deep learning, and the source of all *magic* is always 2. 2 is significantly significantly more powerful. 2 is what surprises you. 2 is when the paddle learns to hit the ball behind the blocks in Breakout. 2 is when AlphaGo beats even Lee Sedol. And 2 is the "aha moment" when the DeepSeek (or o1 etc.) discovers that it works well to re-evaluate your assumptions, backtrack, try something else, etc. It's the solving strategies you see this model use in its chain of thought. It's how it goes back and forth thinking to itself. These thoughts are *emergent* (!!!) and this is actually seriously incredible, impressive and new (as in publicly available and documented etc.). The model could never learn this with 1 (by imitation), because the cognition of the model and the cognition of the human labeler is different. The human would never know to correctly annotate these kinds of solving strategies and what they should even look like. They have to be discovered during reinforcement learning as empirically and statistically useful towards a final outcome.
(Last last thought/reference this time for real is that RL is powerful but RLHF is not. RLHF is not RL. I have a separate rant on that in an earlier tweet
x.com/karpathy/statu…)
Andrej Karpathy@karpathy
DeepSeek (Chinese AI co) making it look easy today with an open weights release of a frontier-grade LLM trained on a joke of a budget (2048 GPUs for 2 months, $6M). For reference, this level of capability is supposed to require clusters of closer to 16K GPUs, the ones being brought up today are more around 100K GPUs. E.g. Llama 3 405B used 30.8M GPU-hours, while DeepSeek-V3 looks to be a stronger model at only 2.8M GPU-hours (~11X less compute). If the model also passes vibe checks (e.g. LLM arena rankings are ongoing, my few quick tests went well so far) it will be a highly impressive display of research and engineering under resource constraints. Does this mean you don't need large GPU clusters for frontier LLMs? No but you have to ensure that you're not wasteful with what you have, and this looks like a nice demonstration that there's still a lot to get through with both data and algorithms. Very nice & detailed tech report too, reading through.
English
Iric retweetledi
I'm a proud Dane. The Danes have done many things very well. But huge advantage in homogeniety. 87% ethnic Danes. Wrote about this fairytale a while back: world.hey.com/dhh/the-realit…
English
@justinspratt This is hilarious. 🤣
I haven't seen dots connected like this since Graham Hancock's last book. 🤣🤣🤣
English
marc is a very good writer, but this was extra special. threading truth with levity is enjoyable reading. worth the (long) read.
Marc Andreessen 🇺🇸@pmarca
*Issue #147 of "The Last Sane Professor" - A Newsletter About Our Inevitable Doom* Dear Fellow Truth-Seekers, [CRITICAL NOTICE: This newsletter has been printed on paper made from hemp grown in a Faraday-caged greenhouse, using ink infused with mycorrhizal fungi to resist quantum surveillance. DO NOT READ ON DIGITAL DEVICES.] From my bunker-office at Stanford (I've reinforced the walls with lead-lined first editions of McLuhan and Debord), I write to you with trembling hands. After three months of sleepless nights analyzing Peter Thiel's blockchain transactions through the lens of Vedic numerology, I've uncovered connections that can no longer be ignored. The release of ChatGPT wasn't just a technology launch – it was a mass initiation ritual. ## The Berlusconi Protocol: Origins of the Current Chaos Let me lay out the evidence chronologically (though time itself may be a construct of Big Tech): 1991: The Soviet Union falls. Simultaneously, Tim Berners-Lee releases the World Wide Web. Coincidence? The same year, a young Elon Musk is reading Douglas Adams, absorbing the idea that reality might be a simulation. Meanwhile, in Italy, Silvio Berlusconi is perfecting the template that would become the playbook for our current techno-oligarchic nightmare. The Berlusconi method: 1. Acquire media control 2. Transform politics into entertainment 3. Blur the line between business and state 4. Deploy bunga bunga technology (more on this in my upcoming paper in the Journal of Memetic Warfare) But here's what everyone missed: Berlusconi wasn't just a businessman turned politician – he was the beta test for a new form of reality manipulation. His "gaffes" weren't gaffes at all, but carefully coded messages to his successor-initiates. ## The Silicon Valley Escalation: Following the Money Through the Looking Glass [The following section was written in a fugue state after discovering that the Stanford particle accelerator's maintenance schedule perfectly aligns with Bitcoin price fluctuations] Let's connect some dots that "they" don't want connected: - 2008: Bitcoin whitepaper released by "Satoshi Nakamoto" (an anagram of "A Samsung Takes Ohio" if you rearrange the letters and add some letters and remove some letters) - 2009: Peter Thiel writes that he no longer believes democracy and freedom are compatible. The same year, Twitter adds the "retweet" button, fundamentally altering human consciousness - 2011: Marc Andreessen declares "software is eating the world." Nobody asks what happens AFTER the software finishes eating - 2012: Elon Musk starts SpaceX. Claims it's for Mars colonization. But what if Mars is just a RED HERRING? (Literally red) - 2016: Trump elected. Immediately starts communicating in the same memetic language as tech billionaires. This is not an accident. - 2020: COVID lockdowns force everyone online. Zoom becomes ubiquitous. Has anyone ever seen Zoom's server farms? ANYONE? - 2021: Facebook becomes Meta. Meta is Hebrew for "death." It's also Greek for "beyond." Death beyond what? THEY WON'T SAY. - 2022: SBF's empire collapses. But was it really an empire, or a massive psycho-social experiment in collective delusion? - 2023: Chat-GPT released. Everyone focuses on AI training data. NO ONE asks about what it's training US to do. - 2024: X (formerly Twitter) introduces neural link integration. Claims it's "just for testing." Testing WHAT? ## The Great Confluence: Why Now? [This section was transmitted via shortwave radio to a network of trusted transcriptionists using a cipher based on defunct cryptocurrency values] Consider these TOTALLY RELATED events from the past month alone: 1. Elon Musk attends UFC events with Trump while wearing a Cybertruck shirt. The geometric patterns in the Cybertruck design, when processed through medieval Islamic geometric algorithms, spell out "SIMULATION BOUNDARY ERROR" in Aramaic. 2. Mark Zuckerberg builds a underground bunker in Hawaii that's allegedly for "family use." But my analysis of the building permits (obtained through FOIA requests filed in palindromic sequences) reveals that its dimensions exactly match the proportions of ancient Mesopotamian ziggurats. 3. Sam Altman's OpenAI saga wasn't a corporate power struggle – it was a ritual enactment of the ancient mystery school tradition of death and rebirth. Notice how he was "fired" for exactly 87 hours? 87 is the atomic number of Francium, the most unstable naturally occurring element. Wake up, people! 4. Bill Gates has been buying farmland across America. Everyone asks why, but nobody asks WHY NOW? I've overlaid his land purchases with ley lines and historical patch notes from Windows 95. The pattern exactly matches the distribution of defunct Radio Shacks. ## The Academic-Industrial Complex: We're All in the Panopticon Now [Written on the back of rejected grant applications] Here at Stanford, I've noticed disturbing patterns in student behavior: - 76% increase in students wearing AirPods during lectures (are they really AirPods, or neural dampeners?) - The university's new "AI Ethics Center" is housed in a building with exactly 1,024 windows (2^10 - a binary message hiding in plain sight) - The campus WiFi password changes follow a pattern that, when plotted against the Fibonacci sequence, predicts future NASDAQ closings with 66.6% accuracy - Our computer science department's coffee machine now requires a blockchain wallet to dispense coffee. I've been surviving on rainwater and fermented kombucha for weeks. ## The Memetic Warfare Phase: We're Already in World War V [This section was composed while wearing a tin foil mortarboard] The real battle isn't happening in any physical domain – it's occurring in what I call the "memetic sphere." Evidence: 1. Billionaires posting increasingly surreal memes at 3 AM: These aren't just sleep-deprived rants, they're reality-warping incantations. I've analyzed them with software designed to detect patterns in whale songs (mammals known for their resistance to techno-capitalist paradigms). 2. Every major tech CEO suddenly becoming obsessed with AI safety: They're not worried about AI; they're preparing us for the revelation that reality itself is a recursive simulation. Think about it: why do they call them "recursive neural networks"? What's being recursed? WHERE DOES THE RECURSION END? 3. The crypto ecosystem's constant creation of new tokens: Each one is a sigil, a piece of geometric magic designed to fragment consensus reality. Why do you think they call it "mining"? They're mining the collective unconscious! ## The Great Tech Reset: What They're Not Telling You [Transcribed from notes hidden in various editions of "Snow Crash" in university libraries across the Bay Area] The endgame is clearer than ever: Phase 1 (Completed): - Replace human interaction with digital dopamine loops - Convince everyone to carry surveillance devices voluntarily - Transform currency into pure information - Make reality itself subscriptionbased Phase 2 (In Progress): - Replace representative democracy with algorithmic governance - Convert all human experience into tradeable tokens - Merge corporate and state power through "regulatory innovation" - Prepare population for "post-scarcity" (read: technofeudal) economy Phase 3 (Incoming): - Reveal simulation hypothesis as fact - Initialize quantum social credit system - Begin great filter event (disguised as "software update") - [REDACTED BY ENTITIES UNKNOWN] ## The Stanford Connection: Silicon Valley's Secret History [Written in invisible ink on the backs of rejected tenure applications] Why did they build Silicon Valley next to major universities? The pattern is hidden in plain sight: - Stanford's particle accelerator forms a perfect phi ratio with Google's quantum computer facility - The original Hewlett-Packard garage is aligned with ancient sacred sites when viewed from satellite - Facebook's first server rack was housed in a building whose address numerologically reduces to the same value as the Antikythera mechanism ## Recent Developments That Cannot Be Coincidental [Documented during a 72-hour caffeine-fueled analysis session] 1. Jack Dorsey's beard growth patterns match the fibonacci sequence 2. Peter Thiel's investment portfolio, when mapped onto star charts, forms the constellation of Ophiuchus 3. WeWork's collapse was actually a mass ritual to prepare commercial real estate for the "great urban reset" 4. The exact number of semiconductors produced in Taiwan last year forms a perfect magic square when expressed in hexadecimal ## A Warning from the Ivory Tower [Transmitted via a network of reformed crypto miners] My academic career may be nearing its end. The dean keeps asking why I've covered my office windows with screenshots of Elon Musk's deleted tweets. My research funding has been cut since I submitted a paper titled "The Metamorphosis of Late Stage Capitalism into Early Stage Simulation Theory: A Technoshamanic Analysis." But I cannot stay silent. The patterns are too clear, the connections too obvious. When the final phase begins, remember: the metaverse isn't just coming – it's already here, and we're living in its beta test. ## Call to Action (If Free Will Still Exists) 1. Switch to analog everything 2. Learn to decode memetic warfare 3. Start a Substack before they're replaced with neural feeds 4. Stock up on hardcover books (they can't be remotely deleted) 5. Question every software update 6. NEVER update your LinkedIn profile during a full moon ## In Conclusion (Though Linear Time is a Construct) [Written in a secure location using a typewriter powered by bicycle generator] If you're reading this, you're either part of the awakening or an AI sentiment analysis algorithm. Either way, remember: When Elon Musk tweets about Mars, he's not just talking about space exploration – he's preparing us for the revelation that reality itself is open source software. I must go now. My Geiger counter is picking up unusual readings from the direction of Apple's new campus, and I need to update my manifesto before the next faculty meeting (which I attend via mirror to avoid facial recognition). *Send help. Or Dogecoin. Or carrier pigeons trained in post-quantum cryptography.* Professor [REDACTED] Department of [REDACTED] Stanford University (This signature has been encrypted using a cipher based on early PayPal source code) P.S. If this newsletter suddenly stops publishing, look for my coded messages in the pattern of Tesla Supercharger installations. P.P.S. To my students: This will definitely be on the final exam, assuming we still exist in this reality by then. --- *Subscribe now to receive my upcoming paper: "Neural Networks or Neurological Control?: A Paranoid Professor's Guide to the Technofeutal Takeover" (pending peer review by a panel of rogue game theorists and quantum mystics)*
English
@mcgregor_rob94 @pigeon_chess I can't fathom the stupidity of these policies.
Every place that has tried rent control has failed (Barcelona is the latest prominent European example).
Alternatively, Tokyo demonstrated the inverse. By increasing supply, it can control its rental costs.
English
I sound like a broken record, but this is needs to be the focus to address rent/housing prices in Namibia: make it easier to increase supply. Hopefully someone in government will finally realise this.
Arpit Gupta@arpitrage
"A 1% increase in new supply (i) lowers average rents by 0.19%, (ii) effectively reduces rents of lower-quality units, and(iii) disproportionately increases the number of second-hand units available for rent. Moreover, the impact on rents is equally strong in high-demand markets”
English