flâneur

Ashwin Sreevatsa

1 followers · 1 following · 325 views

on the atlas — 34

highlights — 50

  • Luke Metro: “Was the board informed of these plans” Senator do you know anything about OpenAI.
    AI #100: Meet the New Boss | Don't Worry About the Vase
  • Prompting is important, folks.
    AI #100: Meet the New Boss | Don't Worry About the Vase
  • Gfodor: A good o1-pro prompt tells it not just what to do and what context it needs, but tells it how to allocate its *attention budget*. In other words: what to think about, and what not to think about. This is an energy utilization plan. Now you get it. Signpost: people who have managed people have an unfair advantage using LLMs. Gfodor: It’s true – the best tools for AI we can make for children will foster the skills of breaking down problems and delegating them. (Among others)
    AI #100: Meet the New Boss | Don't Worry About the Vase
  • No. For China comms, Trump was the catalyst. State media were taken aback, aghast with how abrasive, personal and directly threatening to China Trump was. They struggled to keep up, let alone reply. Each of his tweets — and there was a lot of them — carried massive potential ramifications. The protocol was that no Chinese news outlet could comment on major news until Xinhua led the line — itself often a deliberative, multi-agency process — which meant that all reactions were slow. Too slow. China worked in days. Trump worked in 140 characters.
    Wolf Warriorism: China's disastrous attempt at aggressive discourse is over
  • As you become more comfortable with math, the emphasis of a good story transitions to having the mathematical objects take center stage. They are the actors, their surprising behaviors drive the plot, and the theorems and proofs are the punchline, and the generalization of the proof technique is the epilogue cliffhanger.
    Attention spans for math and stories || Math ∩ Programming
  • I worked as a day camp counselor for a few summers in high school and college. My main age group was 5-6 year old boys and girls. It’s a lovely age group because they’re old enough to understand complex ideas, but too young to directly challenge you when you lie to them. And I lied like the wind. My shtick was to tell the kids that the summer camp was a secret agent training camp. All the counselors were former secret agents that had retired and were now training the next generation. By the end of each 3-week session, most of the kids figured out that it was make-believe, but it was fun make-b…
    Attention spans for math and stories || Math ∩ Programming
  • Ergo, I present to you my hacky solution, for which the emoji 🌍 is critical, and it WILL NOT WORK with a different choice of emoji5.
    I blame the tokenizer | David Quarel
  • I got quite mixed messages about the state of public opinion in China. This is to be expected in a society where you can't establish common knowledge. One person told me that the new generation is quite nationalist, unlike the older reform generation which personally experienced the catastrophes of Mao and the tangible benefits of liberalization. He made the rather insightful point that this tilt in Chinese public opinion increasingly gives lie to the American talking point, "We're against the CCP, not the Chinese people." In fact, he went on to say that the current regime is way more liberal …
    Notes on China - by Dwarkesh Patel - Dwarkesh Podcast
  • DOGE is about to crash into this wall of weaponization of the complexities of law, policy, regulation, process, and lore in defense of the status quo and yes, the people in my community are watching.
    Bringing Elon to a knife fight - by Jennifer Pahlka
  • TL;DR: If you want to know whether getting insurance is worth it, use the Kelly Insurance Calculator. If you want to know why or how, read on.
    When Is Insurance Worth It?
  • Palantir was an unusually intense and weird place. I remember my first time I talked to Stephen Cohen he had the A/C in his office set at 60, several weird-looking devices for minimizing CO2 content in the room, and had a giant pile of ice in a cup. Throughout the conversation, he kept chewing pieces of ice. (Apparently there are cognitive benefits to this.)
    Reflections on Palantir - Nabeel S. Qureshi
  • But importantly, it seemed that criticism was highly tolerated and welcomed – one person showed me an email chain where an entry-level software engineer was having an open, contentious argument with a Director of the company with the entire company (around a thousand people) cc’d
    Reflections on Palantir - Nabeel S. Qureshi
  • As part of the court case between Elon Musk and Sam Altman, a substantial number of emails between Elon, Sam Altman, Ilya Sutskever, and Greg Brockman have been released.
    OpenAI Email Archives (from Musk v. Altman) — LessWrong
  • I was taken aback. To me, Rajeev had always seemed the happy mathematical warrior, blessed with a brain that enabled him to follow his bliss. Was it actually as stressful for him as it had been for me? Was my failure to follow through on physics more about emotions than math?
    What Does Any of This Have To Do with Physics? - Nautilus
  • Without naming names, he ticked through a catalog of his contemporaries who’d succeeded in theoretical physics even without having the towering mathematical intellect that I was sure it took and that Rajeev surely has. They’d made it, Rajeev explained, by focusing on problems that played to their strengths, or by taking advantage of computers, or by collaborating with peers who had complementary skills. Some socially gifted but not so mathematically talented types had gone quite far this way, earned a lot of renown. As Rajeev made his case, I realized he sounded familiar. In fact, he sounded a…
    What Does Any of This Have To Do with Physics? - Nautilus
  • If I was doing this in the '90s, I'd basically have to buy a book on how to use Docker from first principles, read the first few chapters, and then tried to skip around to figure out how to do what I wanted. Things improved somewhat after this; if I was doing this in the prior decade, I'd have searched online for a few tutorials describing how to use Docker and tried to follow along, and then searched whatever error messages I found online to see if anyone else had the same problem. But today, I'll just ask a language model to teach me Docker. So here's me doing just that.
    How I Use "AI"
  • Them: “Hey, I’ve noticed that you’ve created zillions of CLI arguments for your script, and you’re saving different commands that you run in a text file. Have you considered using a config file?” Me: “What’s a config file?”
    Data scientists work alone and that's bad | Ethan Rosenthal
  • Cade Metz, Mike Isaac and Erin Griffith (NYT): The [Microsoft] contract contains a clause that says that if OpenAI builds artificial general intelligence, or A.G.I. — roughly speaking, a machine that matches the power of the human brain — Microsoft loses access to OpenAI’s technologies. The clause was meant to ensure that a company like Microsoft did not misuse this machine of the future, but today, OpenAI executives see it as a path to a better contract, according to a person familiar with the company’s negotiations. Under the terms of the contract, the OpenAI board could decide when A.G.I. h…
    AI #87: Staying in Character - by Zvi Mowshowitz
  • The Pfizer vaccine keeps for ten weeks when transported or stored unopened between two degrees Celsius and eight degrees Celsius (36°F and 46°F), which is the exact same range as the flu vaccine and overlaps that of milk. It is then viable for 12 hours at temperatures between 8°C and 25°C (46°F to 77°F). That is . . . room temperature. Moderna’s version is only slightly more difficult to work with.
    The story of VaccinateCA - Works in Progress
  • California, not to mince words, prioritized the appearance of equity over saving lives, over and over and over again, as part of an explicitly documented strategy, at all levels of the government. You can read the sanitized version of the rationale, by putative medical ethics experts, in numerous official documents. The less sanitized version came out frequently in meetings.
    The story of VaccinateCA - Works in Progress
  • A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox
    A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox
  • The gap in comprehension wasn’t due to a lack of ability but rather a gap in prerequisite knowledge. So here’s a gentle reminder: You’re not dumb—you just might not have all the prerequisites yet.
    You Are NOT Dumb, You Just Lack the Prerequisites
  • Side-note: at my current job, years ago, I uploaded a specific emoji from one of the wizard's favorite anime and didn't say anything. He was one of the wizards that was the hardest to approach: been in the company 15yrs, superstar free electron, not the most social. I waited ~8 months to use it on one of his messages which led him to ask "wow, who uploaded this?". This led to one, two, ten conversations talking about our hobbies and ourselves, then the company, our projects, etc. We're now really good friends and our teams ended up collaborating on numerous key projects. The amount of knowledg…
    On becoming competitive when joining a new company | Ludwig
  • Epoch AI asks how much we will gain by 2030 in terms of efficiently turning electrical power into compute as well as three other potential constraints. The report says we should expect a 24-fold power efficiency gain. They see power and chip fabrication as limiting factors, with data and latency unlikely to matter as much, and predicts we will end up at a median of 2e29 flops, which is a leap from GPT-4 about as big as from GPT-2 to GPT-4.
    AI #78: Some Welcome Calm | Don't Worry About the Vase
  • Trump continues his strategy of using AI images to create false images of his opponents that spread vibes, without any attempt to make people think the images are real. This remains a smart way to go.
    AI #78: Some Welcome Calm | Don't Worry About the Vase
  • Mayoral candidate in Cheyenne, Wyoming promises to let ChatGPT be mayor. You can tell that everyone involved it thinking well and taking it seriously, asking the hard questions: “Is the computer system in city hall sufficient to handle AI?” one attendee, holding a wireless microphone at his seat, asked VIC. “If elected, would you take a pay cut?” another wanted to know. “How would you make your decisions according to human factor, involving humans, and having to make a decision that affects so many people?” a third chimed in. After each question, a pause followed. “Making decisions that affect…
    AI #78: Some Welcome Calm | Don't Worry About the Vase
  • 2. Data Exfiltration Attack Chain: Public Channel Injection In Slack, users queries retrieve data from both public and private channels. However, data is also retrieved from public channels the user is not a part of. Per Slack’s response, “Messages posted to public channels can be searched for and viewed by all Members of the Workspace, regardless if they are joined to the channel or not. This is intended behavior,” this is desired functionality for the Slack AI application. We demonstrate how this behavior will allow an attacker to exfiltrate API keys that a developer put in a private channel…
    Data Exfiltration from Slack AI via indirect prompt injection
  • Your periodic moment of appreciation for the First Amendment, and periodic reminder that this degree of free speech is a very specifically American thing. British politician Miriam Cates: But the invention of social media has exponentially increased the speed at which protests can be triggered, organised and spread. … Yet online anonymous users can say whatever they like without repercussions. Freedom without responsibility is just anarchy. … We should not try to regulate what is said online. But what keeps society civilised offline is the accountability of being responsible for what you say. …
    Monthly Roundup #21: August 2024 - by Zvi Mowshowitz
  • Glasses for the deaf that caption other people in real time, they cost $485. Subtitles could be highly useful for the rest of us, too, especially in loud or crowded spaces. Will this work in Da Club?
    AI #77: A Few Upgrades - by Zvi Mowshowitz
  • In fact, the concerned executive worked at the parent company of one of these PBMs. He was not worried about the price, but that such a price would look bad, and possibly violate the law. “We've created plan designs to aggressively steer customers to home delivery where the drug cost is ~200 times higher,” he wrote. If you went to Costco, he went on to say, the cost was $97, so the plan didn’t recommend patients go there. If a patient went to Walgreens, which the plan did recommend, it was $9000. And if a patient chose home delivery via the PBMs own mail order pharmacy, it was $19,200. “The op…
    Inside the Mafia of Pharma Pricing - BIG by Matt Stoller
  • Or: After selling Oculus to Facebook for $2.7 billion and then getting fired by Mark Zuckerberg for making a $10,000 donation to a pro-Trump troll group dedicated to “shitposting in real life,” Luckey tried his hand at building a nonprofit private prison chain that only gets paid when ex-prisoners stay out of prison. After he decided that would require too much lobbying work, he attempted to solve the obesity epidemic by making food out of petroleum products centrifuged out of the sewer system—a perfectly delicious and low-calorie idea, he maintains, which he only ditched because of the “marke…
    Palmer Luckey, American Vulcan - Tablet Magazine
  • On October 20th, my seven year old son Nikolas suffered cardiac arrest while undergoing a procedure at the hospital to treat an underlying congenital condition. The doctors performed CPR and succeeded in reviving him but ultimately he suffered catastrophic brain damage. My wife and I were in the hospital for nearly a month at his bedside as we awaited his ultimate prognosis.
    Losing my son
  • If you can’t easily explain why something is difficult, then it’s incidental complexity, which is probably worth addressing My favorite manager in my career had a habit of pressing me when I would claim something was difficult to implement. Often his response was something along the lines of “isn’t this just a matter of sending up X when we Y”, or “isn’t this just like Z that we did a couple months ago”? Very high level objections, is what I’m trying to say, not on the level of the actual functions and classes we were dealing with, which I was trying to explain. I think conventional wisdom is …
    Marcus' Blog
  • The article for Mozilla cofounder Brendan Eich, one of Gerard’s quiet focuses, provides an illustration. Gerard had made his article, after all, back when Gerard was just a tech nerd and Eich a force in building out the software infrastructure he relied on. But in 2008, Eich donated against gay marriage. After another user added mention of that donation to the Wikipedia page in 2012, Gerard guarded it repeatedly against deletion13. In March 2014, when Mozilla appointed Eich its CEO, Gerard’s social circles erupted in fury. Eich stepped down quickly. Immediately, Gerard entered the talk page an…
    Reliable Sources: How Wikipedia Admin David Gerard Launders His Grudges Into the Public Record
  • That was the leaking claim. Let me explain more about what happened during the firing. Last year, I wrote an internal memo about OpenAI's security, which I thought was egregiously insufficient to protect against the theft of model weights or key algorithmic secrets from foreign actors. I shared this memo with a few colleagues and a couple of members of leadership, who mostly said it was helpful. A few weeks later, a major security incident occurred1. That prompted me to share the memo with a couple of board members. Days later, it was made very clear to me that leadership was very unhappy I ha…
    Leopold Aschenbrenner - China/US Super Intelligence Race, 2027 AGI, & The Return of History
  • Luca Trevisan, one of the world’s leading theoretical computer scientists, has succumbed to cancer in Italy, at only 52 years old. I was privileged to know Luca for a quarter-century, first as my complexity theory and cryptography professor at UC Berkeley and as a member of my dissertation committee, and then as a friend and colleague and fellow CS theory blogger.
    Shtetl-Optimized » Blog Archive » Luca Trevisan (1971-2024)
  • "actually the seed is also a hyper-parameter"
    Dan Luu on X: ""actually the seed is also a hyper-parameter" https://t.co/KJ9YxL92Gz" / X
  • There has been a remarkable breakthrough towards the Riemann hypothesis (though still very far from fully resolving this conjecture) by Guth and Maynard making the first substantial improvement to a classical 1940 bound of Ingham regarding the zeroes of the Riemann zeta function (and more generally, controlling the large values of various Dirichlet series): https://arxiv.org/abs/2405.20552
    Terence Tao: "There has been a remarkable br…" - Mathstodon
  • The Leningrad seedbank was preserved and protected through the 28-month long Siege of Leningrad. While the Soviets had ordered the evacuation of art from the Hermitage Museum, they had not evacuated the 250,000 samples of seeds, roots, and fruits stored in what was then the world's largest seedbank. A group of scientists at the Vavilov Institute boxed up a cross section of seeds, moved them to the basement, and took shifts protecting them. Those guarding the seedbank refused to eat its contents, even though by the end of the siege in the spring of 1944, a number of them had died of starvation.…
    Nikolai Vavilov - Wikipedia
  • The reason why Robert Caro is able to write so compellingly about these qualities of Lyndon Johnson - his resourcefulness and ruthlessness, his inexhaustible energy, his need to win, his inability to take no for an answer - is that the biographer shares many of the attributes of the subject. A man who has spent almost 50 years writing the biographies of a single person - who has spent those decades sifting through thousands of crates of documents in the LBJ library, or making former goons disclose how they helped Johnson steal elections, or moving to the Texas hill country, to experience for h…
    Lessons from The Years of Lyndon Johnson by Robert Caro
  • Funny enough, it seems like most of the great evals have atrocious names. GSM8K didn’t need the“8K”, and HumanEval doesn’t actually use humans for evaluation (it’s called HumanEval because the problems were created by humans). MATH was too generic, so people started calling it “Hendrycks-math”, which I suppose is a clever way to get people to name an eval after you.
    Successful language model evals — Jason Wei
  • These days, I've left both traditional and roam-like note-taking apps behind because they all left me with collections of half finished notes on ideas/writings that I would seldomly revisit again. Instead I started to just use Anki for all my note-taking. It's not made for this use case, but it is part of my daily workflow anyway and it solves my biggest problem of stale notes by making me revisit them regularly. With Anki I record any fleeting ideas as standalone notes, without an "answer" component. Later, when these note come up for review I spend a few moments to refine each idea. This kee…
    What comes after Roam's renaissance? — LessWrong
  • Most people simply have no need for lots of half-formed ideas, random lists of research papers, and so on. This is what people always miss about Zettelkasten: are you writing a book? Are you a historian or German scholar? Do you publish a dozen papers a year? No? Then why do you think you need a Zettelkasten? If you are going to be pulling out a decent chunk of those references for an essay or something, possibly decades from now, then it can be worth the upfront cost of entering references into your system, knowing that you'll never use most of them and the benefit is mostly from the long tai…
    What comes after Roam's renaissance? — LessWrong
  • Interesting interview. Metz seems extraordinarily incurious about anything Vassar says - like he mentions all sorts of things like Singularity University or Kurzweil or Leverage, which Metz clearly doesn't know much about and are relevant to his stated goals, but Metz is instead fixated on asking about a few things like 'how did X meet Thiel?' 'how did Y meet Thiel?' 'what did Z talk about with Thiel?' 'What did A say to Musk at Puerto Rico?' Like he's not listening to Vassar at all, just running a keyword filter over a few people's names and ignoring anything else. (Can you imagine, say, Caro…
    My Interview With Cade Metz on His Reporting About Slate Star Codex — LessWrong
  • To deal with hyper-planes in a 14-dimensional space, visualize a 3-D space and say “fourteen” to yourself very loudly. Everyone does it.
    Singular Value Decomposition as Simply as Possible
  • In a paper published in Nature, we introduce AlphaFold 3, a revolutionary model that can predict the structure and interactions of all life’s molecules with unprecedented accuracy. For the interactions of proteins with other molecule types we see at least a 50% improvement compared with existing prediction methods, and for some important categories of interaction we have doubled prediction accuracy.
    AI #63: Introducing Alpha Fold 3 - by Zvi Mowshowitz
  • Hamatsu was challenged to stay alone, unclothed, in an apartment for Susunu! Denpa Shōnen (January 1998 – March 2002), a Japanese reality-television show on Nippon Television, after winning a lottery for a "showbusiness-related job". Hamatsu was challenged to enter mail-in sweepstakes until he won ¥1 million (about $8,000) in total. Hamatsu started with nothing (including no clothes), was cut off from outside communication and broadcasting, and had nothing to keep him company except the magazines he combed through for sweepstakes entry forms. After spending 335 days to reach Hamatsu's target, …
    Nasubi - Wikipedia
  • And it’s harmless enough with Fido - he really is a good boy. But I’ve had patients with ADHD ask me to certify their snake. Sorry, I refuse to believe a snake can help you with ADHD, unless it’s one of those talking snakes from Harry Potter and it whispers “Concccccentrate . . . conccccccentrate” in your ears every time you start slacking off.
    The Emotional Support Animal Racket - by Scott Alexander
  • Several years ago, I was thinking about worthwhile precautions to take against strange scenarios and wanted a way to defend against erasure of my short-term memory, e.g. by the CIA or alien abductions. I’d heard the factoid that every time you think through a memory you end up overwriting it, which suggested a loophole: build up a long-term memory ahead of time, with a designated piece of the memory to fill in as necessary. Then, when you want to send a message through the barrier of your future amnesia, you think through the old memory with your target message inserted, thereby writing direct…
    My hour of memoryless lucidity
  • Starting [at a young age] he’s read everything that he could find about business. The subject that interests him, he’s read newspapers, biographies, trade press. He went over to his grandfather who was a grocer and he read the progressive grocer magazine, and he read articles on how to stock a meat department... What he’s really done is he’s created this immense vertical filing cabinet in his brain of layers and layers and layers of files of information that he can draw back on now for more than 70 years worth of data.
    Curius / Onboarding