Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
AI Models Can Provably Hide Arbitrary Capabilities
A research paper arguing that AI models can provably hide arbitrary capabilities from evaluators under formal assumptions, with implications for AI safety evaluations, capability elicitation, and how we interpret benchmark and red-team results.
53 min · 12,294 words
Introducing Strands Decider 2B: a small, open source, decision model
The Strands Agents team releases Strands Decider 2B, an open-source decision model built from a Qwen3.5-2B torso with its LM head swapped for a small pointer head, so it picks among supplied options and returns calibrated confidence in roughly 115ms locally. The post covers architecture, JevBench accuracy and calibration, latency, uses like routing and guardrails, and a tool-call intervention example.
7 min · 1,612 words
Intelligent If-Statements: Jev vs. LLMsWhat if your if-statements could think?
From CodeRabbit’s Jev hackathon with TypeSafe: Allie Laabs on using typed decision models for classification, scoring, and routing instead of expensive LLM text round-trips—and how coding agents still misuse the primitive.
8 min · 1,779 words
Claude-shaped scienceWorking on Claude-shaped problems with BootLoops, a toolkit for exact calculations in quantitative science
Harvard physicist Matthew Schwartz’s Anthropic guest post on BootLoops: an open-source harness that steers LLMs toward checkable, quantitative “Claude-shaped” problems across physics, ecology, genetics, and more—with human experts supplying scientific taste.
5 min · 1,244 words
Why we built the fastest robust TTS model
Gradium's latest streaming TTS hits ~50ms time-to-first-audio while improving naturalness and hard cases like phone numbers—freeing latency budget for LLM turns and barge-in in voice agents.
2 min · 431 words
Surprisingly Complex Waves Reveal the Brain's Inner Workings
Unexpected patterns traveling across the human brain may be reorganizing its activity in real time.
9 min · 2,113 words
The Rise of Overfit Inference Engines
A hands-on essay on how local LLM inference stacks are overfit to specific models and hardware—and what that means for anyone chasing reproducible homelab performance with llama.cpp and friends.
10 min · 2,324 words
Functional ultrasound imaging from scratch
Functional ultrasound imaging (fUSI) is a brain imaging modality that is just starting to be demonstrated in human studies. I’m sure many will be familiar with structural ultrasound imaging, which forms static images using ultrasound, as in a sonogram1. Functional ultrasound instead captures movies of brain tissue very rapidly, tracking subtle changes caused by the blood flow that follows neural activity.
11 min · 2,451 words
Can you use autoregressive diffusion to generate market data?
A Jane Street 2026 summer intern project explores generative models for order-book market data, treating book events as both continuous and discrete, comparing DDPM with flow matching, handling discontinuities and atom smoothing, and examining the resulting rollouts.
8 min · 1,737 words
AutoBenchmark: benchmark creation & the role of humans
*Figure 1. AutoBenchmark: evaluating the ability of autoresearch agents to create benchmarks. AutoBenchmark instructs an autoresearch agent to create a benchmark, we specifically target benchmarks that evaluate autoresearch agents themselves (i.e., benchmarking autoresearch). Within the benchmark creation loop, the autoresearch agent iteratively revises the benchmark via two main sources of feedback: one from the benchmark solvers (including the same model used as the benchmark creator) and the other from external verifiers of the benchmark (AI or human). We thus also…
16 min · 3,723 words
Tokenization: A Survey for Modern NLP
A comprehensive survey by 32+ tokenizer researchers covering subword algorithms, multilingual pitfalls, evaluation, theory, security, and alternatives—framing tokenization as a first-class bottleneck in the LLM era.
2 min · 376 words
Doing a Machine Learning PhD While Working in Japan
Marco Cognetta recounts completing a CS PhD at Tokyo Tech on MEXT while working part-time at Google: visa flexibility, the three-year clock, name-recognition tradeoffs, lab life, and whether to stay in Japan afterward.
8 min · 1,867 words
Gemini 4 Argon: our next era of frontier intelligence
Today, we’re announcing our new frontier model, Gemini 4 Argon, which is rolling out to a set of trusted cyber defenders through our Fairwind Program. Built to sustain deep reasoning across complex, longhorizon workflows, Argon is fundamentally changing the way we work and build at Google.
6 min · 1,378 words
Comparing Muon, NorMuon and AdamW for Fine-tuning a Dense Retriever
Qingcheng Zeng gives Muon and NorMuon the same tuning budget as AdamW when fine-tuning a contrastively pretrained dense retriever: lower training loss, no BEIR win. Learning rate and transfer matter more than the optimizer.
5 min · 1,151 words
Language Models for Text Classification: From Bag-of-Words to JevA visual guide to bag-of-words, RNNs, CNNs, transformers, Jev-like APIs, and calibration
Sebastian Raschka walks from classic bag-of-words classifiers through RNNs, CNNs, and transformers to TypeSafe AI's Jev—explaining APIs, IMDb benchmarks, calibration, and why decision models matter for agent harnesses.
5 min · 1,076 words
CLMs treat context as an editable file so models natively manage their own context—beating harness baselines on BrowseComp-Plus and long-horizon coding with fewer FLOPs, plus in-context skill evolution and RL for context management.
3 min · 699 words
Hugging Face’s Tarek Ziadé explains Serge, a CI agent that finds Transformers failures, reproduces them on GPUs, writes patches, verifies them, and opens PRs—29 merges in ~80 days.
9 min · 2,070 words
Harness Learning Enables Generalizable Test-Time Adaptation
arXiv paper introducing harness learning: a proposer model revises an agent’s executable harness from execution feedback, improving test-time adaptation on unseen reasoning and multi-hop QA without weight updates.
56 min · 12,765 words
Anthropic introduces Claude Sonnet 5.5, a faster and lower-cost complement to Opus 5.5 that improves agentic coding and everyday task performance versus Sonnet 5.
8 min · 1,747 words
Build an LLM Tokenizer and Attention from Scratch in TypeScript
SitePoint tutorial that implements BPE tokenization, cosine similarity vector search, and scaled dot-product attention in TypeScript—exposing LLM primitives as ordinary readable code.
25 min · 5,827 words