Laerdon Kim

on the atlas — 102

• Investigating Model Editing for Unlearning in Large Language Models1 savers
• [2610.05534] Dataset Signatures in Human-LLM Interactions and User Modeling1 savers
• From Sampling to Reinforce7 savers
• Language Models use Lookbacks to Track Beliefs1 savers
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges6 savers
• John Hewitt · Language & Machine Learning1 savers
• Lab Policy on AI in Writing & Communication · John Hewitt3 savers
• Destroy Any Website - Sprite Fusion5 savers
• On the Value of Doing a PhD in the Age of AI12 savers
• [2604.21691] There Will Be a Scientific Theory of Deep Learning3 savers
• [2507.06445] Can Interpretation Predict Behavior on Unseen Data?1 savers
• [2609.17933] AI Mediators Regulate Emotion and Create Value in Disputes2 savers
• Advice to the New College Student in the AI Era3 savers
• Neil Rathi4 savers
• Opinion | If You’re Over 40, You’re Ready to Use A.I. - The New York Times2 savers
• [2607.20734] LLMs Get Lost in Evolving User Intent1 savers
• SWE-chat: Coding Agent Interactions From Real Users in the Wild4 savers
• How do you use a different theme in a remote VS Code window - Stack Overflow1 savers
• [2605.09808] Quantifying the Utility of User Simulators for Building Collaborative LLM Assistants1 savers
• microgrants! - Google Docs2 savers
• The Truth Lies Somewhere in the Middle (of the Generated Tokens)1 savers
• [2601.17087] Lost in Simulation: LLM-Simulated Users are Unreliable Proxies for Human Users in Agentic Evaluations1 savers
• Lindia Tjuatja1 savers
• Schedule · AI Engineer World's Fair 20261 savers
• Introduction | RLHF and Post-Training Book by Nathan Lambert1 savers
• Representation Engineering: A Top-Down Approach to AI Transparency5 savers
• Alane Suhr2 savers
• Notes on the Industry Job Search18 savers
• Research:Newsletter/2026/May - Meta-Wiki1 savers
• PhD_thesis_Shirley_Wu_final.pdf2 savers
• HumanLM1 savers
• Today's AI Talks Like “Nobody.” New Research Gives It Real Personality. | Stanford HAI1 savers
• [2604.07629] Behavior Latticing: Inferring User Motivations from Unstructured Interactions1 savers
• The Limitations of Large Language Models for Understanding Human Language and Cognition | Open Mind | MIT Press1 savers
• ML Job Interviews: The Ultimate Guide – Silvia Sapora11 savers
• [2505.12540] Harnessing the Universal Geometry of Embeddings1 savers
• [2606.06624] Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory1 savers
• [2512.05648] Beyond Data Filtering: Knowledge Localization for Capability Removal in LLMs1 savers
• Join Geodesic Research and Help Build the Base for Alignment1 savers
• [2510.26745] Deep sequence models tend to memorize geometrically; it is unclear why2 savers
• [2509.24653] Identity Bridge: Enabling Implicit Reasoning via Shared Latent Memory1 savers
• [2602.15029] Symmetry in language statistics shapes the geometry of model representations3 savers
• [2604.21691] There Will Be a Scientific Theory of Deep Learning5 savers
• Reasoning as Trajectories1 savers
• The Simple Habit That Saves My Evenings | alikhil | software engineering, kubernetes & self-hosting2 savers
• Find the stable and pull out the bolt – Natalie B. Hogg1 savers
• Nathan Godey1 savers
• user_interactions.pdf1 savers
• [2508.19227] Generative Interfaces for Language Models1 savers
• [2510.05056] Modeling Student Learning with 3.8 Million Program Traces1 savers
• [2511.04427] Speed at the Cost of Quality: How Cursor AI Increases Short-Term Velocity and Long-Term Complexity in Open-Source Projects1 savers
• Andi Peng2 savers
• Crawl entire websites with a single API call using Browser Rendering · Changelog1 savers
• [2602.15829] Operationalising the Superficial Alignment Hypothesis via Task Complexity1 savers
• Tiny TPU10 savers
• Group | Sherry Tongshuang Wu1 savers
• Checklists Are Better Than Reward Models For Aligning Language Models - Apple Machine Learning Research1 savers
• 2025.acl-long.453.pdf1 savers
• NeurIPS Invited Talk The Art of (Artificial) Reasoning1 savers
• Paloma3 savers
• the html review 049 savers
• How Does A Blind Model See The Earth? - by henry2 savers
• On the Tradeoffs of SSMs and Transformers | Goomba Lab8 savers
• the-illusion-of-thinking.pdf7 savers
• Explore | alphaXiv2 savers
• Jean-Rémi KING2 savers
• EC05-Chen.dvi12 savers
• the friendship theory of everything69 savers
• An Opinionated Guide to ML Research59 savers
• 95%-ile isn't that good44 savers
• escaping flatland: career advice for CS undergrads43 savers
• The Persona Selection Model: Why AI Assistants might Behave like Humans36 savers
• Daylight Computer Co.30 savers
• THE 2028 GLOBAL INTELLIGENCE CRISIS26 savers
• A Gentle Introduction to Graph Neural Networks26 savers
• Notes on Roadtrips26 savers
• Why We Think | Lil'Log26 savers
• Things I learned in college | Kat Huang25 savers
• Lil'Log24 savers
• The machines are fine. I'm worried about us.23 savers
• AI as Normal Technology | Knight First Amendment Institute22 savers
• Aman's AI Journal • Primers • Ilya Sutskever's Top 3021 savers
• What Is ChatGPT Doing … and Why Does It Work?—Stephen Wolfram Writings19 savers
• My Career as a Series of Emails18 savers
• The Unreasonable Effectiveness of Recurrent Neural Networks18 savers
• Neural Networks, Manifolds, and Topology -- colah's blog16 savers
• Motherfucking Website16 savers
• Thinking Machines Lab11 savers
• Part 1: Key Concepts in RL — Spinning Up documentation11 savers
• Failing to Understand the Exponential, Again10 savers
• What Does Any of This Have To Do with Physics? - Nautilus10 savers
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learning8 savers
• mymind is the extension for your mind.8 savers
• The Best GPUs for Deep Learning in 2023 — An In-depth Analysis7 savers
• anhedonia | writing6 savers
• At 17, Hannah Cairo Solved a Major Math Mystery | Quanta Magazine6 savers
• Magic UI5 savers
• The Paris Review - Loving the Limitations of the Novel: A Conversation between Sally Rooney and Merve Emre - The Paris Review4 savers
• you have all these rules, and you think they'll save you4 savers
• Overflowing with thoughts; at a loss for words3 savers

highlights — 33

• I have been using this trick for more than 5 years now, and it helps me to keep my work and life balanced. Here are the two main ideas of it: Don’t overwork Write down the next steps before finishing your workday

  The Simple Habit That Saves My Evenings | alikhil | software engineering, kubernetes & self-hosting

• more generally i think the only real way to make progress on your problems is by doing things you love and losing yourself in action. it must be done through joy, not grief

  you have all these rules, and you think they'll save you

• more generally i think the Box looks something like: you have some big-picture goal for how you want to be (eg. making good music) but the big-picture goal remains in the background while some proxy goal (eg. reading and playing sheet music) occupies your immediate attention. the proxy and big-picture goals are correlated but misaligned - taking steps towards the proxy gets you closer to the big-picture goal (especially when you’re just getting started), but optimizing too hard for the proxy can take you further from where you actually want to be. failure to realize your goals are proxies can …

  you have all these rules, and you think they'll save you

• Registers are named rax, rbx, rcx, rdx, rdi, rsi, rbp, rsp, and r8-r15 for a total of 16 of them. The “letter” ones are named like that for historical reasons: rax is “accumulator,” rcx is “counter,” rdx is “data” and so on — but, of course, they don’t have to be used only for that.

  Assembly Language - Algorithmica

• We do not address Sense Gain and Sense Loss Detection, as they are relatively novel formulations.

  Lexical Semantic Change through Large Language Models: a Survey | ACM Computing Surveys

• Neuro-Symbolic Concepts: Representations

  Jiayuan Mao

• beware of old people with money who want to use you as a missile. as an undergrad, your career path is maximally flexible and your skillset is minimally specialised: you’re a missile that can in principle be pointed at any technical problem. many older people have strong monetary incentives to point you at problems they care about, and to convince you to do things that may not be for you but rather for them.

  escaping flatland: career advice for CS undergrads

• To convert a tail-recursive method to a while loop, we wrap the method body in a big while loop with guard true, then replace each recursive call with assignment statements that reassign to the method's formal parameters the new values that would have been passed in the recursive call. Here is what the binary search code looks like after this transformation:

  Object-Oriented Design and Data Structures

• int sum(Node n) { int s; for (s = 0; n != null; n = n.next) s += n.data; return s; }

  Object-Oriented Design and Data Structures

• For recursive code to be correct, the base case of the recursion must eventually be reached on every chain of recursive calls.

  Object-Oriented Design and Data Structures

• In this assignment, there is nothing special about the first row of a table. Although it will often be used as a header, containing labels for each column, our simplified join operation will always look for matches in the first column whether or not the first entries in those columns are the same.

  Assignment 3: Merging Spreadsheets (CS 2110 Fall 2024)

• remember that asserts are for catching programming bugs, not user errors

  Assignment 3: Merging Spreadsheets (CS 2110 Fall 2024)

• An empty list has a null head and a null tail. A list of size 1 has head == tail. A list of size 2 has head != tail, but no nodes in between. A list of size 3+ has at least one node in between head and tail.

  Assignment 3: Merging Spreadsheets (CS 2110 Fall 2024)

• You should not call get() in a loop over indices anywhere in your submission. That pattern is inefficient, because each call to get() will traverse the list from the beginning instead of from where the previous call left off.

  Assignment 3: Merging Spreadsheets (CS 2110 Fall 2024)

• This means that, whenever mutating the list, it is possible that the head pointer, tail pointer, and size will all need to be updated to maintain the class invariant.

  Assignment 3: Merging Spreadsheets (CS 2110 Fall 2024)

• But it is (almost) equivalent to say that the methods from the superclasses are copied down to their subclasses, except when overridden.

  Object-Oriented Design and Data Structures

• filtering out words that fall below a term frequency threshold. While this approach can remove a majority of false positives, it risks the introduction of false nega- tives

  Statistically Significant Detection of Semantic Shifts using Contextual Word Embeddings

• no existing methods attempt to char- acterize the uncertainty of the estimated semantic shift for each individual word

  Statistically Significant Detection of Semantic Shifts using Contextual Word Embeddings

• While previously the industrial revolution and the introduction of machinery in factories dramatically shifted the composition of the US economy and labor market from being one based largely on manufacturing to one based on services (a move which simultaneously lowered effective wages)

  Artificial Intelligence, Automation, and Class Struggle — On The Line

• One particularly striking shortcoming of the A/0 features is that they don't describe what happens when A/0/0 emits a token like href, which leads to a more complex state.

  Towards Monosemanticity: Decomposing Language Models With Dictionary Learning

• Implication Creation (IC) φ ⇒ (ψ ⇒ φ) Implication Distribution (ID) (φ ⇒ (ψ ⇒ χ)) ⇒ ((φ ⇒ ψ) ⇒ (φ ⇒ χ)) Implication Reversal (IR) (¬ψ ⇒ ¬φ) ⇒ (φ ⇒ ψ)

  Introduction to Logic - Lesson 4

• because it essentially means their tax bills won’t rise as fast as the market value of their homes.

  NYC Property Tax Bills Are Seen as Unjust and Opaque. That Could Change. - The New York Times

• Together, we walked through her methodology for conducting interviews and analyzing her qualitative data, before ​“coding” excerpts from her interviews.

  Inequality | Tang Institute at Andover

• What the curb-cut effect demonstrates, writes Blackwell, is that accommodation is not a zero-sum game: “[W]hen we create the circumstances that allow those who have been left behind to participate and contribute fully—everyone wins.”

  The Yale Law Journal - Forum: Beyond the Public Square: Imagining Digital Democracy

• Event Argument Extraction

  2305.03514.pdf

• According to their degrees of salt requirements, halophiles are classified into three groups: slight (0.34–0.85 M salt), moderate (0.85–3.4 M salt), and extreme halophiles (3.4–5.1 M salt) [2].

  Halophile - an overview | ScienceDirect Topics

• Efforts are further needed to utilize the true potential of extremozymes, which is generally possible with the discovery of novel halophilc members and the subsequent screening, optimization and purification strategies.

  Halophile - an overview | ScienceDirect Topics

• hello! i'm JESS ENG, a food writer based in new york city.

  about · jess eng

• One incredibly useful exercise I’ve found is to watch myself program. Throughout the week, I have a program running in the background that records my screen. At the end of the week, I’ll watch a few segments from the previous week. Usually I will watch the times that felt like it took a lot longer to complete some task than it should have. While watching them, I’ll pay attention to specifically where the time went and figure out what I could have done better.

  95%-ile isn't that good

• This is a paper about Harvard reinventing the wheel of computer ethics education which has a long and comprehensive history stretching back to the 1980s

  Embedded EthiCS: Integrating Ethics Across CS Education | August 2019 | Communications of the ACM

• convert them into spectrograms

  Audio Deep Learning Made Simple: Sound Classification, Step-by-Step | by Ketan Doshi | Towards Data Science

• Skilled volunteer/s develop and deliver the data product (for free or at a subsidized rate).

  Why “data for good” lacks precision. | by Sara Hooker | Towards Data Science

• proactively sends her suggestions for music clubs at school.

  Dot by New Computer