Training transformers in seconds, not hours
Someone built a 22KiB transformer that trains in 13 seconds. The trick is not the model size, it is how they squeezed PyTorch out of the picture.
Read the note 1 min read
From the notebook
machine-learning.
41 notes on this topic.
Seven LLMs running in your browser tab, no server involved
Someone built a lab where you can test tiny language models that run entirely client-side. No API calls, no backend, just WebGPU and a lot of patience.
Read the note 2 min read
Chat templates are why models say 'as an AI'
A new paper shows the system prompt is not what makes models say 'as a language model'. It is the chat template wrapper that turns raw output into self-aware hedging.
Read the note 1 min read
Google wants to run ML training in orbit
Project Suncatcher is Google's proposal to deploy GPU clusters in low Earth orbit. The pitch is simple: space is cold, solar power is free, and latency to ground stations is acceptable for batch jobs.
Read the note 1 min read
Someone built an LLM leaderboard that knows what you can afford
A live benchmark that ranks models by performance per dollar spent, not just raw scores. Turns out the best model depends on your budget.
Read the note 1 min read
Mercury 2.5 LLM clocks 770 tokens per second
A new language model is generating text faster than most tools can render it. The bottleneck just moved from the API to the browser.
Read the note 1 min read
The JEV prompting pattern works on models that never heard of it
A Reddit user discovered that giving a model explicit instructions to behave like JEV (a fictional expert persona) produces cleaner outputs, even when the model has no training data about JEV.
Read the note 1 min read
Someone is archiving deleted LLM models before they vanish
Pirate Face archives open-weight language models that companies try to delete. Turns out model takedowns happen more often than I thought.
Read the note 1 min read
Someone fitted neural textures using evolution strategies instead of backprop
A graphics engineer got neural texture maps working with ES optimisation. No gradients, no autodiff, just mutation and survival.
Read the note 1 min read
Transformers without loops: the Inception architecture drops self-attention
A new transformer variant ditches the attention loop entirely. Faster inference, cleaner scaling, same performance.
Read the note 1 min read
Someone built a browser playground for ML-specific programming languages
A live playground for experimenting with domain-specific ML languages, no install required.
Read the note 1 min read
Security cameras that identify birds by sound
Someone repurposed their home security setup to log every bird species that flies past. The microphone was already there.
Read the note 1 min read
Nvidia just bought Hugging Face for thirteen billion dollars
The GPU monopoly now owns the model zoo. This changes who controls open-weight AI.
Read the note 1 min read
Etched built an ASIC that only runs transformers
A startup ditched everything GPUs do well and made a chip that runs one architecture 20 times faster.
Read the note 1 min read
Your local LLM is smarter than you think
Running a model on your laptop does not mean it is broken. The context window might be.
Read the note 1 min read
Claude wrote a printer driver from scratch
An LLM reverse-engineered a 2008 HP laser printer's protocol and generated a working macOS driver. No human debugging required.
Read the note 1 min read
llama.cpp finally ships a v0.1.0
Three years after first commit, the project that made local LLMs possible ships a stable release.
Read the note 1 min read
What happens when you train an LLM on fifth-grade material only
Researchers built an LLM using exclusively elementary-level text. The results challenge assumptions about what makes a model fluent.
Read the note 1 min read
Flock Safety added guardrails after every cop in America bought their cameras
The company that sells license plate readers to 5000 police departments just announced privacy controls. Timing raises questions.
Read the note 1 min read
You can extract the thinking process from closed LLMs
Researchers figured out how to steal the internal reasoning steps from proprietary models like GPT-4o and Claude through API timing attacks.
Read the note 1 min read
Someone fit a reasoning model into 14MB
Needle2 runs on phones, smartwatches, and Raspberry Pis. The entire model is smaller than a single photo.
Read the note 1 min read
Three models are better than one
Why my intrusion detection capstone needed an Isolation Forest, a Random Forest and an autoencoder to reach 0.95 F1 on real attack traffic.
Read the note 2 min read
Training LLMs on an 8GB GPU is now a weekend project
Someone just released a minimal codebase for supervised fine-tuning, direct preference optimisation, and group relative policy optimisation on consumer hardware.
Read the note 1 min read
What LLM labs actually hire for
Research engineer roles at the big labs filter for production ML skills first, paper count second.
Read the note 1 min read
Cisco released a family of small security models that actually run locally
Antares models are 1B to 8B parameters, fine-tuned on security tasks, and Apache 2.0 licensed. This is not another rebranded Llama wrapper.
Read the note 1 min read
Someone built a neural network entirely in SQL
A developer trained MNIST digit recognition using only SQL queries. No Python. No frameworks. Just recursive CTEs and window functions.
Read the note 1 min read
When your production model gets a deprecation notice
Google announced Gemini 2.5 Flash will be discontinued in February 2027. Developers who built on it are scrambling.
Read the note 1 min read
LLM burnout is real
The moment you realise you are debugging prompt chains instead of writing code.
Read the note 1 min read
Mean-field theory might be lying about how neural networks actually learn
Physicists found that the standard mean-field approximation for neural networks breaks down when you look at correlations between neurons.
Read the note 1 min read
Haystack is what happens when you need RAG pipelines that actually run in prod
Deepset's Haystack framework caught my eye because it treats retrieval-augmented generation as a data pipeline problem, not a chatbot wrapper problem.
Read the note 2 min read
Neural Particle Automata: teaching particles to self-organise
Researchers trained particles to form complex shapes without central control. Each particle runs the same neural network, learns local rules, and the swarm organises itself.
Read the note 1 min read
Neural cellular automata running in your browser at 60fps
WebGPU makes training tiny neural networks that grow patterns possible in real-time, no server required.
Read the note 1 min read
Europe wants to train frontier models on a mesh of university GPUs
A GitHub repo sketches how Europe could pool scattered compute across universities and research labs to train a GPT-4 class model without buying a new datacenter.
Read the note 1 min read
Transformer attention has a working memory problem
Researchers found that transformer attention mechanisms lack the executive control functions that let human brains manage working memory. The models can retrieve information, but they cannot suppress irrelevant context.
Read the note 1 min read
Data2prompt stuffs entire projects into LLM context windows
A CLI tool that flattens your data science repo into one massive prompt. Smart filtering meets the 200K token era.
Read the note 1 min read
Norway built an LLM training cluster on 2 PB of Huawei flash
The Norwegian University of Science and Technology is running LLM workloads on Chinese storage hardware. The performance numbers are interesting.
Read the note 1 min read
Content moderation just became a federal requirement
A bill requiring platforms to moderate content encouraging violence against Jewish communities turns moderation from a platform choice into a legal obligation.
Read the note 1 min read
AI bug hunters are spamming the Linux security list into chaos
Linus Torvalds says automated vulnerability scanners have turned the kernel security mailing list into noise. The tools work, the signal-to-noise ratio does not.
Read the note 1 min read
DeepSeek-V4-Flash revives the steering vector idea
Steering vectors let you nudge a model's behaviour without retraining. They fell out of favour when newer models stopped responding to them. DeepSeek-V4-Flash brought them back.
Read the note 1 min read
Someone built a transformer in Rust using category theory
A programmer rewrote PyTorch's transformer architecture using Rust and abstract algebra. The result is dense but fast.
Read the note 1 min read
AI found a CVE in its own training set and we are pretending that is a win
Mythos discovered a vulnerability that was already documented in the data it was trained on. The industry is calling this autonomous discovery.
Read the note 1 min read