Today we're releasing Laguna S 2.1, our most capable model to date. It's a 118B total parameter MoE model with 8B activated per token, a context window of up to 1M tokens, and thinking and no-thinking modes. Small enough to run on a single @NVIDIAAI DGX Spark poolside.ai/blog/introdu...

Introducing Laguna S 2.1

Today we’re releasing Laguna S 2.1, a significant step forward in our development of models that pursue longer horizon work and make effective use of reasoning.

poolside.ai

Today we’re publishing the technical report behind Laguna M.1 and Laguna XS.2. This report opens up more of what went into them: Model Factory, pre-training data, distributed training, post-training, agent RL, quantization, and evaluation. poolside.ai/assets/lagun...

Poolside is hosting a 2-day model research hackathon in London Join us to push an open-weight agent model as far as you can. RL and fine-tune Laguna XS.2, our latest-generation model, on Prime Intellect Lab. Dates: May 29–30 Partners: #nvidia + #PrimeIntellect + @hf.co Prize: NVIDIA DGX Spark ⤵️

As agents get more clever, so do their attempts at benchmark hacking. Last Monday, we found one of our RL runs jumped ~20% on SWE-Bench-Pro over a weekend, reaching ~64% which would make it #1 on the leaderboard. These were our findings: poolside.ai/blog/through...

Through the looking glass of benchmark hacking

Outlining some of the reward hacks we’ve encountered and what strategies we are exploring to resolve them.

poolside.ai

Today we’re shipping Laguna M.1 and Laguna XS.2 – our first public models. We’re also shipping our agent harness and a preview product experience. Both models were trained from scratch on our own stack: data pipelines, training infrastructure, and agent RL. poolside.ai/blog/introdu...

Introducing Laguna XS.2 and Laguna M.1

We’re releasing two foundation models and two products into preview today.

poolside.ai

you cannot compete with someone who is having fun

You would *hope* that software developer job interviews and software development wouldn't require two distinctly different sets of skills.

I clicked a link and ended up on sourceforge.net - AMA

We're still looking for talent across the board, be it in inference, training, backend or IDEs. Feel free to ping if you want to know more. jobs.ashbyhq.com/poolside?utm...

poolside Jobs

poolside Jobs

jobs.ashbyhq.com

ikea captcha

Why don't we include this on every website? So true.

Need a dev advocate? Look no further, Mark is a rock star

Mark ⚡️Heckler@mkheck.bsky.social · last yr.

Just got caught up in the latest round of terminations from Microsoft. If anyone needs a "batteries included" advocate who really knows code, AI, and real production concerns + solutions, ping me! Ready to make great things happen for you. Onward and upward!

branch names often make it obvious how stupid a bug was: bmuskalla/stopAtEnd

how you know you're still a startup? someone just mentioned "long term" and "2 weeks" in the same sentence

Excited to announce the official launch of our online book, WebAssembly from the Ground Up! 🎉 It's the book we wish we'd had 3 years ago. No messing with tools and frameworks. It's a hands-on guide to the core of Wasm: the instruction set and module format. Link below. 👇

Reminder: if you haven’t yet read “The Mythical Man Month,” buy two copies so you can read it faster.

my personal AGI benchmark is implementing atproto oauth from scratch. ultimate evidence of beyond-human-level intelligence

Saving papers or articles to read? Just mark them as stale every morning; last 3 things I pulled from my "to read list" were outdated already. Sigh Anything interesting you got that I can mark as stale by tomorrow?

New prompt technique: stop being smarty pants. nobody wants code comments for indecipherable code. write readable code instead. thank you.

E_TOKENIZER_GONE_ROGUE

OH: a Jupyter notebook is just a dressed up REPL 😂

It's funny how people think that AI tools out there can just implement a "model switcher" and be independent of the underlying model behaviour...brace yourselves. SOTA doesn't mean equal

this s1 paper is so refreshing! Love seeing ML breakthroughs that don't need massive compute or complex tricks 🙌 Clean, simple, and it actually works? More of this please! github.com/simplescalin...

GitHub - simplescaling/s1: s1: Simple test-time scaling

s1: Simple test-time scaling. Contribute to simplescaling/s1 development by creating an account on GitHub.

github.com

I don’t care how hacky it feels but I really dig `uv run —with=“dep>1,<2” myscript.py docs.astral.sh/uv/guides/sc...

Across various programming languages I work with, LLMs became more reliable in fixing imports properly than language tooling itself 🤔

Tiny TIL: I just figured out how to run pytest with a different Python version against my pyproject.toml/setup.py projects using uv run uv run --python 3.12 --with '.[test]' pytest til.simonwillison.net/pytest/pytes...

Programming languages: "We are just a way to operate computers in a way that makes sense to humans." Programming languages [takes a big joint hit]: "What if there were 5 kinds of nothingness?"

When you just thought you‘ve seen everything in the hype cycle. Germany goes

We’re building a new static type checker for Python, from scratch, in Rust. From a technical perspective, it’s probably our most ambitious project yet. We’re about 800 PRs deep!