Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Meta FAIR introduces RL-XAR (Reinforcement Learning from eXpert-Aligned Rubrics): learn rubrics from the gap between expert writing and model output, then train models toward expert-level text generation to reduce AI slop.
16 min · 3,572 words
Can a Model Learn New Skills as Add-Ons?
Connito Research trains residual MoE experts with their own routers on a frozen DeepSeek-V2-Lite base, then merges independently trained math, code, medical, law, and finance experts in seconds without retraining—lifting domain benchmarks while leaving the original model untouched.
4 min · 913 words
On the Value of Doing a PhD in the Age of AI
MIT's Phillip Isola offers ten reminders for anxious AI PhD students: public research still has leverage, human expertise remains safety infrastructure, and the PhD's job is to chase a moving frontier for the love of the game.
3 min · 781 words
Small Decisions: Engineering a Leading Model
AWS engineer Marc Brooker recounts building and training a small leading model hands-on—what worked, how it performed, and what the exercise taught him about modern model-building.
9 min · 2,024 words
The systems that no one will test
The systems that no one will test This happened to me in 2020, and it has been on my mind again lately. During the worst of the pandemic, I found a vulnerability in a system that gave me access to the Brazilian federal system, and with that access I was able to retrieve information on any Brazilian (think of 200+ million people data).
4 min · 901 words
Throughout the history of AI, open research has played a critical role in driving progress. Today, many key details of frontier large language models (LLMs) remain proprietary, but open-weights model families— such as DeepSeek, Kimi, and MiMo —continue to provide a valuable window into the development process for modern LLMs.
7 min · 1,609 words
AMD to Acquire World Labs to Advance the Future of AI Compute
AMD announces a definitive agreement to acquire Fei-Fei Li’s World Labs, bringing spatial AI researchers and models in-house to shape future AI hardware, software, and systems.
4 min · 1,011 words
World Labs announces a definitive agreement to join AMD, arguing that scale, reach, and closer hardware partnership are needed to accelerate spatial and physical AI research.
1 min · 218 words
Robert W argues LLM confidence scores are vibe-words, not calibrated probabilities—and outlines RLCD-style approaches that treat hallucination as a solvable measurement and training problem.
7 min · 1,560 words
Turn GLM-5.3-Flash into a Jev-like System One model
Johannes Hötter walks through turning GLM-5.3-Flash into a fast Jev-like “System One” decision model—typed options with probabilities in a single forward pass, matching Jev’s accuracy and speed.
11 min · 2,426 words
“As a Language Model…”: Chat Template Switches LLM Self-Referential Voice
Research showing chat templates act as a switch between disclaimer (“I’m just an AI”) and experiential (“I feel”) self-referential voices across 8 instruct models, with a steerable activation direction that reproduces the template effect.
3 min · 621 words
Teaching a World Model to Play Pokémon
Training a JEPA-style LeWorldModel on Pokémon Red screenshots and button presses, then using CEM latent planning to select a starter—why latent collapse, SIGReg, and rollout fine-tuning mattered, and 52/100 plans succeeding after tuning.
2 min · 574 words
Some short musings on the shape of language models, e.g. what it means to design a language model around a harness, and not the other way around.
8 min · 1,810 words
DeepSeek Elastic Compute (DSec)
# Computer Science > Distributed, Parallel, and Cluster Computing [Submitted on 19 Sep 2026] # Title:DeepSeek Elastic Compute (DSec): A Sandbox Infrastructure for Effective Agentic Training at Scale View PDF HTML (experimental) Abstract:Large-scale agentic training and evaluation with large language models (LLMs) rely on isolated, stateful execution environments in which models inspect repositories, invoke tools, execute commands, and interact with task-specific services. These workloads create sandboxes in large bursts, span heterogeneous functionality and isolation…
16 min · 3,693 words
Mixture of Experts (MoE) for Backend Engineers
A detailed visual guide to token routing, expert batching, weighted combination, and the memory and communication tradeoffs of MoE serving. Suppose a model has dozens of feed-forward subnetworks, but each token uses only two of them. The arithmetic per token can stay modest while the total weight set grows. Now place that model on eight GPUs. If every GPU stores all experts, memory can become the limit; if experts are split across GPUs, token activations must travel to whichever GPU owns their…
15 min · 3,515 words
Exploding variance of means of exponentials: least-squares to the rescue
Francis Bach reframes log-sum-exp / KL estimation as a continuum of least-squares problems with closed-form spectral solutions—cutting exploding exponential variance.
2 min · 464 words
What happens when you analyze college football like the CIA?
What happens when you analyze college football like the CIA? A couple weekends ago, the Illinois football team lost to Duke at home, 31–27.
11 min · 2,518 words
I Look, if you are still stuck on “AI cannot really think, it’s just a stochastic parrot”, please snap out of it and lock in, or you’ll keep repeating that line until you find yourself sitting in the corner chair, watching as ChatGPT™ has sex with your wife.
8 min · 1,856 words
V-JEPA: Learning Video Representations by Feature Prediction
A hands-on walkthrough of Meta’s V-JEPA: video tubelets, feature prediction, pretrained encoder features, temporal tests, and a small action-classification experiment.
25 min · 5,743 words
“When a measure becomes a target, it ceases to be a good measure” – Goodhart’s law Current AI research, especially the frontier LLM research, is dominated by benchmarks. It is the first thing we look at when a new model comes out, it is the headline of each release, and they dominate the discourse when […]
13 min · 2,940 words