Akira Yoshiyama

on the atlas — 443

• fuz1 savers
• Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment3 savers
• Introducing Beam: Reflection’s 501B open-weight model — Reflection2 savers
• A scientific approach to improving model performance  |  Machine Learning  |  Google for Developers2 savers
• extending swe-2's reward function to steer the pareto frontier, Anish Lakkapragada1 savers
• Marlos C. Machado1 savers
• David Tao1 savers
• The 4-bitter Lesson | humans&1 savers
• adam-maj/tiny-gpu: A minimal GPU design in Verilog to learn how GPUs work from the ground up ·5 savers
• Keeping Your Options Open is Ruining Your Life - by megan17 savers
• Clare Lyle | Notes for my EEML talk on continual learning1 savers
• Clare Lyle | The state of plasticity in 20252 savers
• infosketch.pdf1 savers
• [2506.02285] Why Gradients Rapidly Increase Near the End of Training1 savers
• Tangentially / A Machine Learning Blog | Aaron Defazio, Machine Learning Researcher & Data Scientist1 savers
• Helium Browser1 savers
• Law of the handicap of a head start4 savers
• Destroy Any Website - Sprite Fusion5 savers
• We’re gonna need a lot more mathematicians | What's new3 savers
• pytransitions/transitions: A lightweight, object-oriented finite state machine implementation in Python with many extensions ·1 savers
• Do the Real Thing - Scott H Young12 savers
• mertkahyaoglu/atlas: A reading app for system design interviews: concept modules, technology pages and worked designs with interactive architecture diagrams. ·1 savers
• Hacking CDDVDBlu-ray for fun and scientific research.pdf - Google Drive2 savers
• Oblique Strategies1 savers
• Why Momentum Really Works9 savers
• [2205.14571] Provable Benefits of Representational Transfer in Reinforcement Learning1 savers
• AI Infrastructure at Periodic – Periodic Labs5 savers
• underscores - electric cleaners mix | hyperpop | EC108 - YouTube1 savers
• How to Unclench | Jonny Miller23 savers
• mazoure20a.pdf1 savers
• The Somalia Affair and the Transformation of Canadian Military Justice1 savers
• f514cec81cb148559cf475e7426eed5e-Paper.pdf1 savers
• Progressive Point Matching1 savers
• Preston Fu1 savers
• Magic5 savers
• Anthropic principle4 savers
• [2604.11791] A Mechanistic Analysis of Looped Reasoning Language Models1 savers
• [2606.05555] Representation Learning Enables Scalable Multitask Deep Reinforcement Learning1 savers
• How to Beat Procrastination - LessWrong11 savers
• How our data shaped neural architecture discovery, and how automation can reshape the future | Core Automation5 savers
• CSAPP_2016.pdf2 savers
• [1709.06560] Deep Reinforcement Learning that Matters2 savers
• Deep RL Bootcamp - Lectures3 savers
• Computer Architecture: A Quantitative Approach1 savers
• Microsoft Word - CACM_roofline_complete.doc1 savers
• The Best Textbooks on Every Subject - LessWrong28 savers
• How to Measure Anything — LessWrong3 savers
• Statistical Rethinking2 savers
• Seeing Theory1 savers
• Home | Gentle Systems6 savers
• Research acceleration: The view inside OpenAI | OpenAI14 savers
• GPT-6 Astra on robot arms | Robocurve1 savers
• Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design5 savers
• On solving the Jane Street Reverse Engineering Challenge | jestoph’s tech blog1 savers
• Tips for self-study and independent research | Andy Konwinski2 savers
• OpenAI7 savers
• a11ce ⨳3 savers
• pguo-PhD-grind.pdf4 savers
• Time complexity of operations on built-in types — Python 3.16.0a0 documentation1 savers
• Some Scientists Have ‘Magic Hands’ in the Lab. This A.I. Is Learning Why. - The New York Times2 savers
• DEF CON: The Documentary - 720p - YouTube1 savers
• bunnie's blog1 savers
• Layman's Guide to IC Reverse Engineering3 savers
• Semi-invasive attacks - A new approach to hardware security analysis1 savers
• .:: Phrack Prophile on Gera ::.2 savers
• Horace He6 savers
• HOPE 26 :: pretalx1 savers
• What happens when a GPU reads memory | Doubleword1 savers
• A Gallery of Methods Beyond RL — Part I: Sampling Methods | Shengyu Feng2 savers
• Home | Learning Mechanics DeCal1 savers
• Mountain Biking Packages and Guided Biking Tours in BC1 savers
• Strong Inference: Certain systematic methods of scientific thinking may produce much more rapid progress than others4 savers
• GRPO++: Tricks for Making RL Actually Work2 savers
• Li Jiang | On-Policy Distillation: Promise, Pitfalls, and Prospects1 savers
• Liu Ziyin2 savers
• turnip idioms - musings2 savers
• 24 Hours of Lemons - Wheel to Wheel Racing for $500 Cars3 savers
• Incentives are for losers - by Adam Mastroianni5 savers
• lina.sh1 savers
• TBPN.CA - Public Broadcasts2 savers
• How FPGAs work, and why you'll buy one2 savers
• [2608.13335] Neural Quadratic Forms: A Unified Minimal Model for Sudden Learning and Scaling Laws1 savers
• Is a PhD on Language Models Worth it in 2025?1 savers
• Sturdy Statistics1 savers
• Maximizing the value of your Claude Code sessions | Claude by Anthropic1 savers
• Dan McKinley :: Choose Boring Technology10 savers
• Data vis do’s & don’ts | Datawrapper Blog3 savers
• HIAS1 savers
• The social AI hypothesis - by Stefano Viel - Ste2 savers
• bunnie's blog1 savers
• | Kibotronics1 savers
• Flipping read-protect bits: Altera EP900 | Kibotronics1 savers
• The Unreasonable Effectiveness of Mathematics (EP Wigner).pdf1 savers
• The Future Of Knowledge Workers In An AI-Driven World10 savers
• xoreaxeaxeax/asm-hall-of-shame: Racing to the bottom of CPU performance ·1 savers
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges5 savers
• What will more intelligence actually do for us?3 savers
• Pavlov's List: A List of RL Environment Startups7 savers
• twenty one | arielle lok6 savers
• Fast · Patrick Collison48 savers

highlights — 22

• "he's intense" — everyone. would I live any other way?

  About - Aidan Smith

• This boosts the signal out of Waterloo, and companies compete fiercely for Waterloo interns and grads above all other schools.

  The Last Technical Interview. Today we will pour one out for the… | by Steve Yegge | May, 2026 | Medium

• Above the LLM abstraction sit agentic loops

  How to Land a Frontier Lab Job

• Below the LLM stack sit kernels.

  How to Land a Frontier Lab Job

• The way out of this is to work at the edges of where frontier labs operate

  How to Land a Frontier Lab Job

• never turned up my nose at “menial” work. If you found the right place to dig for gold, just keep digging. Don’t complain about the dirt.

  How to Land a Frontier Lab Job

• "You keep on learning and learning, and pretty soon you learn something no one has learned before."

  career - What's a mathematician to do? - MathOverflow

• Remember, no man is a failure who has friends.

  It's a Wonderful Life

• Yu Sun would like to thank his PhD advisor Alyosha Efros for the insightful advice of looking at the pixels when working on machine learning.

  One-Minute Video Generation with Test-Time Training

• Jevons paradox, the tendency for the more efficient use of a resource to increase total consumption of that resource, rather than decrease it

  Why I’m not worried about AI job loss - David Oks

• short of that corner-solution world, we’re still in a regime governed by human bottlenecks

  Why I’m not worried about AI job loss - David Oks

• labor substitution is about comparative advantage, not absolute advantage.

  Why I’m not worried about AI job loss - David Oks

• Not being firm with when to let go. To let go of people, roles, projects, ideas. Early on, I struggled with knowing when to draw the line: when to stake out vs when to move on. This is yet another highly situational intuition that you only learn by feeling/doing. I've clung to dying initiatives, waning relationships, and outdated self-conceptions well past their expiration dates.

  To What Degree? - Jolie's field notes

• Human consciousness as a brief, lossy interface between the navigation app and the steering wheel.

  How to live without your phone - by Sam Kriss

• Navigating via phone is the opposite: you know exactly where you are, but it’s not really anywhere in particular. You sink into a kind of abstract mental tunnel until you reach where you’re going. A new, inverted form of being lost.

  How to live without your phone - by Sam Kriss

• “Finishers are always monsters”—that is, behind every completed oeuvre is a trail of broken promises to people you love.

  🌻 tryhard - by Jasmine Sun - @jasmine’s substack

• At the heart of the plan is Rapidus, a little-known company backed by the government and some of Japan's biggest corporations including Toyota, Softbank and Sony.

  Japan's gamble to turn island of flowers into global chip hub

• he wasted his entire life, my mom said to me, the evening we found the love letters. his entire life, and mine as well.

  after my dad died, we found the love letters | Jenneral HQ

• Enable config options that optimize for repos with many files in the working directory. With many files, commands such as git status and git checkout may be slow and these new defaults improve performance:

  Git - git-config Documentation

• If a couple aunts and uncles stopped by, you had a social gathering without even trying.

  The Death of Partying in the U.S.A.—and Why It Matters

• If summoning these magnificent technologies incurs the death of our social lives, a permanent surge of anxiety, and the long-term demise of deep friendships, then we’ll have built ourselves a glittering dungeon of insularity and called it progress. Surely, we can do better than that.

  The Death of Partying in the U.S.A.—and Why It Matters

• I know that some augmentations can also be amputations

  The Death of Partying in the U.S.A.—and Why It Matters