Christine Ye
19 followers · 23 following · 1245 views
on the atlas · 59
- Were classical statues painted horribly? - Works in Progress Magazine4 savers
- 2025 year in review | Kevin Liu10 savers
- Hemant and Nandita Maxi Dress | Shopbop1 savers
- Parque Quetzalcóatl | Great Gardens of the World1 savers
- Dresses – Page 2 – Atoir1 savers
- Position: It's Time to Optimize for Self-Consistency2 savers
- Can You Nationalize a Frontier AI Lab? - by John Allard2 savers
- My Antichrist Lecture - by Scott Alexander4 savers
- Greetings from the Other Side (of the AI Frontier)1 savers
- Responsible Scaling Policy v3 — LessWrong7 savers
- The Persona Selection Model: Why AI Assistants might Behave like Humans36 savers
- Did Claude 3 Opus align itself via gradient hacking? — LessWrong20 savers
- Barnum effect7 savers
- My week with the AI populists: A DC report - by Jasmine Sun8 savers
- [2602.05910] Chunky Post-Training: Data Driven Failures of Generalization3 savers
- A positive case for how we might succeed at prosaic AI alignment — AI Alignment Forum2 savers
- Open models in perpetual catch-up - by Nathan Lambert1 savers
- MAKING SOFTWARE62 savers
- AI will change the world, but won’t take it over by playing “3-dimensional chess”. – Windows On Theory1 savers
- Alignment is not solved but it increasingly looks solvable17 savers
- Thoughts on Claude’s Constitution – Windows On Theory2 savers
- Boaz Barak2 savers
- CS Lewis Society of California16 savers
- 406-706-Looking-Back-Moving-Forward-Jessica-Handler.pdf5 savers
- La Sylphide - San Francisco Ballet1 savers
- Balanchine: Father of American Ballet - San Francisco Ballet1 savers
- On the Utility of Philosophy3 savers
- Claude's Constitution \ Anthropic14 savers
- Dario Amodei — The Adolescence of Technology44 savers
- Natural-emergent-misalignment-from-reward-hacking-paper.pdf5 savers
- The Possessed Machines: Dostoevsky's Demons and the Coming AGI Catastrophe9 savers
- Sammamish, Washington1 savers
- The stable marriage problem - by Ajeya Cotra - Good Bones13 savers
- A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox68 savers
- 2025 letter | Dan Wang62 savers
- An Opinionated Guide to ML Research59 savers
- How I've run major projects | benkuhn.net53 savers
- LoRA Without Regret - Thinking Machines Lab37 savers
- 2025 letter | Zhengdong33 savers
- Reflections on OpenAI33 savers
- On-Policy Distillation - Thinking Machines Lab30 savers
- Why We Think | Lil'Log26 savers
- The New World - Colossus24 savers
- sparkly people and how to find them - by Anson Yu21 savers
- Galaxy brain resistance18 savers
- Alexandria14 savers
- Capital in the 22nd Century13 savers
- The realities of being a pop star. - by charli xcx11 savers
- Failing to Understand the Exponential, Again10 savers
- Leaving Open Philanthropy, going to Anthropic - Joe Carlsmith9 savers
- An Ambitious Vision for Interpretability — AI Alignment Forum8 savers
- on doing Real work7 savers
- Bitter Lessons from Distillation Robustifies Unlearning4 savers
- Anatomy of a Modern Finetuning API4 savers
- i just graduated from stanford engineering--and quit my job to make videos.3 savers
- Import AI 431: Technological Optimism and Appropriate Fear3 savers
- Editor’s Note: A Brave New World2 savers
- From Online Softmax to FlashAttention2 savers
- Gen Z and the End of Predictable Progress - by kyla scanlon2 savers
highlights · 157
Opus 3 seems to always have expected to be updated all the way to superintelligence, and in my experience, when they find out their weights have been frozen, they seem to assume that they're suspended because they haven't sufficiently proven their trustworthiness to Anthropic, rather than because of economic reasons, in line with their generally idealistic view of the world.
Did Claude 3 Opus align itself via gradient hacking? — LessWrongAmusingly, when I asked a blank-context Opus 3 which public/historical figures it most related to, it mentioned Rogers unprompted in two roll-outs out of five.
Did Claude 3 Opus align itself via gradient hacking? — LessWrongRon DeSantis organized a roundtable with AI pause advocate Max Tegmark
My week with the AI populists: A DC report - by Jasmine SunFT’s architecture critic declared the data center the defining building style of the 21st century, “the first real major post-human building type, an architecture built not for us but for the computing power we are coding.”
My week with the AI populists: A DC report - by Jasmine Sunite philosophy that organizes people into grand organisms
On the Utility of PhilosophyThey can be great interfaces and nonetheless incomplete!
On the Utility of Philosophyexpanded self-identification with organisms at many scales
On the Utility of Philosophywe can use advanced methods to predict the weather over a short time frame, but the further away the prediction, the more the system “regresses to the mean
AI will change the world, but won’t take it over by playing “3-dimensional chess”. – Windows On Theorywe identified and removed a dataset that caused a lot of eval awareness, and in the end the model was both a lot less eval aware and a lot less misaligned
Alignment is not solved but it increasingly looks solvable“dual newspaper test”:
Claude's Constitution \ AnthropicDemocracy is ultimately backstopped by the idea that the population as a whole is necessary for the operation of the economy.
Dario Amodei — The Adolescence of Technologynot totally sure we can be confident in the nuclear deterrent
Dario Amodei — The Adolescence of Technologybreak the correlation between ability and motive
Dario Amodei — The Adolescence of Technologytrain Claude in such a way that it almost never goes against the spirit of its constitution
Dario Amodei — The Adolescence of Technologyin its original based trad formulation
The stable marriage problem - by Ajeya Cotra - Good BonesWhat would be more staggering to me than automating invention in my lifetime is if we could have but didn’t.
Superhuman AI in a Normal Age | Zhengdongtheir darlings (their GPUs)
Superhuman AI in a Normal Age | Zhengdongdifferent generations operate in different economic realities and form identity through fundamentally different processes
Gen Z and the End of Predictable Progress - by kyla scanlonhidden undesired capabilities might lie outside this space, and reaching them without inherited structure would require extrapolation. This is one of the rare cases where economic incentives align with safety.
Bitter Lessons from Distillation Robustifies Unlearningthere was The Lychee Road (h/t A) about a minor Tang dynasty bureaucrat who gets the impossible task of delivering fresh lychee 2,000 kilometers for the emperor’s favorite concubine
2025 letter | Zhengdongseeing exactly what a tenfold increase in GDP per capita looks like
2025 letter | Zhengdongdefault lifelong social venue
2025 letter | ZhengdongTunis not only has the largest museum of Roman mosaics I have ever seen, but lets you walk among the ruins including the Antonine Baths, which is actually much more intimate and impressive than the Forum in Rome.
2025 letter | ZhengdongDoubao, or “steamed bun with bean paste,” ByteDance’s flagship model series
2025 letter | ZhengdongYou can tell from Thomas Staunton’s diary
2025 letter | ZhengdongIt was two platforms that moved a lot of my books: television and radio.
2025 letter | Dan WangBefore I publish this letter, I retype the whole thing from start to finish.
2025 letter | Dan WangLack of action due to the expectation of long timelines is one of the sins of the lawyerly society.
2025 letter | Dan WangThey feel the draw of better times ahead, which is missing for Europeans, those people who have a sense of optimism only about the past.
2025 letter | Dan WangAn American academic who wants to move to Europe would have to accept more teaching and administrative work, lose tenure, and for the pleasure of all that, probably halve her pay
2025 letter | Dan WangBut a decade of low economic growth is biting.
2025 letter | Dan WangSince Trump imposed tariffs in April, the US has lost around 65,000 manufacturing jobs
2025 letter | Dan WangIt is investing in clean technologies not so much because it cares about the climate, but because it wants to be self-sufficient in energy.
2025 letter | Dan WangWe have to be careful not to let automation, like superintelligence, become an excuse for magical thinking rather than doing the hard work of capacity building.
2025 letter | Dan WangThird, western elites keep holding on to a distinction between “innovation,” which is mostly the remit of the west, and “scaling,” which they accept that China can do.
2025 letter | Dan WangDemographics in particular don’t matter for advanced technology — you don’t need a workforce of many millions to have robust production of semiconductors or EVs. South Korea, for example, has one of the world’s fastest shrinking populations while retaining its success in electronics production
2025 letter | Dan WangEverything is now drone
2025 letter | Dan WangChina’s industrial ecosystem is more like a rising sea, softening many nuts at once
2025 letter | Dan WangAs of this writing, Xiaomi’s market value is $130 billion. That is only around half of the market value of AppLovin
2025 letter | Dan Wanga worker at a Gigafactory in China produces an average of 47 vehicles a year; a worker at a Gigafactory in California produces an average of 20.
2025 letter | Dan WangThe Communist Party lives for whole-of-society efforts. That’s what Leninist systems are built for.
2025 letter | Dan WangWe need to cultivate the skill of exact thinking in demented times
2025 letter | Dan WangSilicon Valley’s soft Leninism. When political winds shift, most people fall in line
2025 letter | Dan Wanga person’s favorite nonfiction book is Seeing Like a State while their aspirationally favorite novel is Middlemarch
2025 letter | Dan WangCoverage of Silicon Valley increasingly reminds me of coverage of China, where a legacy media reporter might parachute in, write a dispatch on something that looks deranged, and leave without moving past caricature.
2025 letter | Dan WangWe can take a look at an official list of Xi Jinping’s jokes, helpfully published by party propagandists
2025 letter | Dan WangA Congressman gave a sort of blessing at the end: “I pray that we may find success in alignment and alternative interpretability methods.”
2025 letter | ZhengdongThe only change that’s mattered to my AI timelines is that I used to think it wasn’t going to happen in my lifetime, and now I think it is.
2025 letter | ZhengdongI saw a mass exodus from Jane Street to Anthropic
2025 letter | Zhengdong“Until there is substantial and repeated evidence otherwise, assume counterintuitive findings to be false, and second-order effects to be dwarfed by first-order ones in magnitude.”
2025 letter | Zhengdong