19
1
A Treatise on How to Use the Internet Without Committing Philosophical Suicide (pastebin.com)
4
Are all evils caused by insufficient knowledge? (bretthall.org)
2
Leverage Research 1.0 (lydialaurenson.substack.com)
1
TreeSheets (Hierarchical Spreadsheet) (strlen.com)
21
Why eval startups fail (2025) (thomasliao.com)
2
Are You in the Weights? (intheweights.com)
3
Cursor and SpaceX: In search of a complete loop (kwokchain.com)
35
Zen and the Art of Machine Learning Research (jxmo.io)
1
Demystifying Noise Contrastive Estimation (jxmo.io)
1
Joan Didion: Staking Out California (1979) (nytimes.com)
3
Gram Newton-Schulz: A Fast, Hardware-Aware Newton-Schulz Algorithm for Muon (tridao.me)
1
The Decline of Token-Level Purchasing Power (bigspin.ai)
2
A one-parameter model that gets 100% on ARC-AGI-2 (eitanturok.github.io)
2
The Effective Sample Size (smola.org)
2
Can you go 82-0? (82-0.com)
1
Automating Plain-Text Location Updates with Apple Shortcuts and Redis (nanjiangwill.com)
6
Bias Compounds, Variance Washes Out (convergentthinking.sh)
55
Rothko for your current weather conditions (joonas.wtf)
1
When AI Starts Writing Systems Code (coreauto.com)
1
DTensor, Correctness and the Costs of Abstraction (runwayml.com)
137
The just-say-no engineer was a ZIRP phenomenon (seangoedecke.com)
2
The AI Decoupling (vintagedata.org)
9
The Lottery – Shirley Jackson (1948) (newyorker.com)
2
Yagni (martinfowler.com)
5
LoRA and Weight Decay (2023) (irhum.github.io)
9
Through the looking glass of benchmark hacking (poolside.ai)
1
Diversity as the Bottleneck in Self-Play (ivison.id.au)
3
The Moth Story Map (themoth.org)
1
Boosting multimodal inference performance by >10% with a single Python dict (modal.com)
20
Why are neural networks and cryptographic ciphers so similar? (2025) (reiner.org)
5
Using group theory to explore the space of positional encodings for attention (janestreet.com)
1
Pyptx – A Python DSL to Write Nvidia PTX for Hopper and Blackwell (github.com/patrick-toulme)
11
Which one is more important: more parameters or more computation? (2021) (parl.ai)
1
As We May Think (theatlantic.com)
30
Work with the garage door up (2024) (andymatuschak.org)
18
Six (and a half) intuitions for KL divergence (perfectlynormal.co.uk)
2
Thoughts on AI and Research [pdf] (economics.mit.edu)
1
How fast can algorithms advance capabilities? (epoch.ai)
3
An Example of Statistical Investigation of the Text Eugene Onegin – Markov, 1913 [pdf] (alpha60.de)
3
Proactively Parasocial (nicklandolfi.com)
111
Ripgrep is faster than grep, ag, git grep, ucg, pt, sift (2016) (burntsushi.net)
6
Fast regex search: indexing text for agent tools (cursor.com)
1
Just One More Prompt (Peter Steinberger) (steipete.me)
1
In high school I found a mistake in the Martian, DM'd Andy Weir, and he replied (twitter.com/zachary_horvitz)
1
BullshitBench (petergpt.github.io)
1
AI as Normal Technology [pdf] (amazonaws.com)
13
Book: The Emerging Science of Machine Learning Benchmarks (mlbenchmarks.org)
2
NanoGPT Speedrun Living Worklog (tylerromero.com)
2
Neolab and Emerging AI Lab Tracker (cleverhack.com)
1
Pplx-Embed: Embedding Models for Web-Scale Retrieval (perplexity.ai)
16
The Pleasures and Pains of Coffee (1830) (umich.edu)
2
The Flying Wallendas (wikipedia.org)
3
Escaping flatland: career advice for CS undergrads (ong.ac)
2
Forge: Scalable Agent RL Framework and Algorithm (minimax.io)
14
Continuous batching (2025) (huggingface.co)
1
Continual learning and the post monolith AI era (baseten.co)
8
Why is everything so ugly? (nplusonemag.com)
1
DoorDash and Other Food Delivery Apps Are Reshaping Mealtime (nytimes.com)
1
Future leakage in block-quantized attention (matx.com)
2
The Normalization of Deviance in AI (embracethered.com)
4
Bertrand Russell on Apricots (1935) (peterhousehold.blogspot.com)
1
Clawdbot Remembers Everything (twitter.com/manthanguptaa)
39
Exactitude in Science – Borges (1946) [pdf] (kwarc.info)
2
Learning with LLMs (jwuphysics.github.io)
26
Slouching Towards Bethlehem – Joan Didion (1967) (saturdayeveningpost.com)
2
LLM architecture has evolved from GPT-2 to GPT-OSS (2025) (modal.com)
11
Weight Transfer for RL Post-Training in under 2 seconds (perplexity.ai)
16
Keeping 20k GPUs healthy (modal.com)
1
The State of LLM Serving in 2026: Ollama, SGLang, TensorRT, Triton, and vLLM (thecanteenapp.com)
1
What Is Arithmetic Bandwidth? (modal.com)
4
GPU memory snapshots: sub-second startup (2025) (modal.com)
2
Teenygrad (github.com/tinygrad)
1
Meaning in Large Language Models: Form vs. Function (jgehring.net)
1
Shape Suffixes – Good Coding Style (medium.com/noamshazeer)
4
Building Internal Agents (lethain.com)
1
How to Use LLM as a Judge (Without Getting Burned) (twitter.com/manthanguptaa)
42
[flagged] FTX whistleblower Caroline Ellison set for early release next month (invezz.com)
3
Tips for Writing a Technical Book (borischerny.com)
1
Information, complexity, brains and reality (Kolmogorov Manifesto) (2007) (arxiv.org)
4
Dinosaur Food: 100M year old foods we still eat today (borischerny.com)
50
Two kinds of vibe coding (davidbau.com)
2
Statistical Learning Theory and ChatGPT (kamalikachaudhuri.substack.com)
7
I ran out of money, spent my savings on a Hong Kong prostitute,& became a commie (docs.google.com)
5
Why are your models so big? (2023) (pawa.lt)
1
[dupe] The Decline of Deviance (experimental-history.com)
87
Sycophancy is the first LLM "dark pattern" (seangoedecke.com)
2
Contextualization Machines (stochasm.blog)
1
ChaCha has all the answers – unless I'm on the other end (2009) (archive.org)
8
What I don’t like about chains of thoughts (2023) (samsja.github.io)
1
Reframing Impact (turntrout.com)
1
Continuous Batching from First Principles (huggingface.co)
1
Solving Kilordle (hauntsaninja.github.io)
1
Kilordle (jonesnxt.github.io)
1
What makes good reasoning data (huggingface.co)
2
Evaluating the Effectiveness of LLM-Evaluators (a.k.a. LLM-as-Judge) (eugeneyan.com)
1
Compute Forecast (AI 2027) (ai-2027.com)
1
A Realistic AI Timeline (vintagedata.org)
2
Biotech companies I wish existed (eladgil.com)
1