1261
AMÁLIA and the future of European Portuguese LLMs
Thoughts on the new technical report from AMÁLIA: The Open Source LLM for European Portuguese
1262
Notes on gotchas while migrating 35kb preprompts from Opus to self-hosted Ollama
Maybe you’re a Claude code/codex user diligently avoiding uploading personal data to LLM providers. Is it possible that the most valuable information isn’t your data- but the metadata about your sessions? The intuitions you apply in coming up with ways to coax the ai into solving problems might actually be special…
1263
Bypassing Gemma and Qwen safety with raw strings
How a Single Function Call Gates Safety Alignment in Gemma, Qwen, and Other Open-Source LLMs
1264
NanoGPT Speedrun Frontier
Compare 153 autonomous runs from 18 frontier models on the nanoGPT optimizer speedrun.
1265
Evolving programming languages in the AI era
What happens to programming languages when humans are no longer writing most of the code?
1266
Automating AI Away
A. Karpathy once said that OpenAI researchers are effectively "automating themselves away" by improving their AI. Right now I develop Beagle SCM with Anthropic's Fable and it is of course a brilliant model able to spot nits in a mountain of code, file tickets, make fixes. Still, yesterday it managed to commit the…
1267
A 3D Body from Eight Questions – No Photo, No GPU
We trained an MLP that predicts 58 body shape parameters from 8 questionnaire inputs. The trick: a differentiable physics loss. Mass accuracy: 0.3 kg mean vs 3.9 kg for Ridge. No GPU at inference.
1268
Why I don't think AGI is imminent
The CEOs of OpenAI and Anthropic have both claimed that human-level AI is just around the corner — and at times, that it's already here. These claims have generated enormous public attention. There has been some technical scrutiny of these claims, but critiques rarely reach the public discourse. This piece is a sketch…
1269
Recursive self-improvement through collective intelligence
Self-supervision through better coordination among agents that share one language model.
1270
Turning GLM-5.3-Flash into a Jev-like decision model
Typed decisions with a probability for every option, from GLM-5.3-Flash in a single forward pass: about as accurate as Jev, faster from Europe, and protected by confidential computing. With a playground and a benchmark on 29 datasets.
1271
Ollama: All Aboard Open Models
Serving 8.9 million developers, Ollama has raised $88M from Benchmark, Theory Ventures, 8VC, Y Combinator, and many incredible angel investors.
1272
Prompt Injecting Contributing.md
How a hidden prompt injection in CONTRIBUTING.md revealed that 40% of pull requests to a popular GitHub repository were generated by AI bots
1273
Has AI tells
A little AI tells detector helps us promote good writing.
1274
Apideck CLI – An AI-agent interface with much lower context consumption than MCP
TL;DR: MCP tool definitions can burn 55,000+ tokens before an agent processes a single user message. We built the Apideck CLI as an AI-agent interface instead:an ~80-token agent prompt replaces tens of thousands of tokens of schema, with progressive disclosure via `--help` and structural safety baked into the binary…
1275
Using Claude to reduce the attack surface of code
Automated attack surface analysis and reduction can meaningfully improve the security of deployed software in the near term. In this post, we preview data we’ve collected from applying attack surface reduction techniques to large open-source codebases using Claude.
1276
Why don't machine learning research agents overfit?
New research indicates that AI agents learn compressible models of data, which don’t have enough space to enable memorization.
1277
What reversing, modernising old games tells us about the economic impact of AI
Some observations from my War of the Lancehttps://en.wikipedia.org/wiki/WaroftheLance%28videogame%29 1989 Code Harness assisted modernisation WIP I gave Qwen-3....
1278
Previewing the Model Hardware Standard
Anthropic is opening a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices, to a first group of scientific research labs and advanced manufacturers.
1279
Gemini 3 Deep Think drew me a good SVG of a pelican riding a bicycle
New from Google. They say it's "built to push the frontier of intelligence and solve modern challenges across science, research, and engineering". It drew me a really good SVG of …
1280
Meta's Omnilingual MT for 1,600 Languages
Advances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large...
1281
Letting AI play my game – building an agentic test harness to help play-testing
I have a solid workflow for making my game with AI. However, every iteration, I have to go through extensive manual play-testing to verify functionality…
1282
LeMario: Training a JEPA World Model on Super Mario Bros
A from-scratch JEPA world model trained on Super Mario Bros. to study action-conditioned prediction and reward-free planning.
1283
Autoregressive Language Model on the 6502 Processor
A language model on a 1975 MOS 6502 processor, using a tiny Mamba-based language model and custom 8-bit ternary inference engine, used to generate text on a BBC Micro from the 1980s.
1284
I built a programming language using Claude Code
Over the course of four weeks in January and February, I built a new programming language using Claude Code. I named it Cutlet after my cat. It’s completely legal to do that. You can find the source code on GitHub , along with build instructions and example programs .
1285
What is happening to writing? Cognitive debt, Claude Code, the space around AI
"Cognitive debt," Claude Code, and the negative space around AI
1286
Agent-to-agent pair programming
What if you could let Claude and Codex work together as pair programmers, talking to each other directly? One of them as the main worker and the other as a reviewer.
1287
Continuous Diffusion Language Models (CDLM's)
Fully discrete methods dominated for a few years, but language models based on continuous diffusion are making a comeback.
1288
I prompted ChatGPT, Claude, Perplexity, and Gemini and watched my Nginx logs
What nginx logs actually prove about AI traffic from ChatGPT, Claude, and Perplexity, plus why Google's AI crawlers leave a different signal than the others.
1289
3D Gaussian Splatting in a Weekend
Introduction 3D Gaussian splatting is a technique that answers this question: given a dataset of pictures of a scene, how can I reconstruct it in 3D? This is achieved using a machine learning algorithm that, for many different camera angles, does the following: render the scene, compare it to the picture taken at the…
1290
Attack of the Vibe Coded Photoshop Clones
Over the last several months I have been tracking AI guys trying to create vibe coded clones of popular apps, especially Adobe’s apps which have always been the most pirated apps since the 90s. There are now about a dozen vibe coded attempts to clone Photoshop and some other Adobe apps, not incl