Rasmus Ros

Constrained black-box optimization for production ML. PhD (continuous opt). Building COMBO at Kotlin libs on the side.

[Rasmus Ros @rosrasmus]

Hi, I'm Rasmus. Head of engineering by trade, building Eignex on the side: open-source infrastructure for optimization loops in production.

Posting here as I build it.

https://t.co/NNicwJtYdI

#buildinpublic #opensource

[Rasmus Ros @rosrasmus]

Anthropic: the worlds most dangerous model too powerful to release.
Also Anthropic: we have no idea why we are forced to pull Fable. It's just a misunderstanding.

[Rasmus Ros @rosrasmus]

The best part about new Anthropic model releases is the accompanying weekly limit resets from all the bug fixes.

[Rasmus Ros @rosrasmus]

Cool article about optimizing images in RAG pipelines. Precomputing image-to-text is obvious, but they have some nice extensions that are worth reading about. https://t.co/AuSohdQtcH

[Rasmus Ros @rosrasmus]

Spent the afternoon watching Claude confidently diagnose a Kotlin/Wasm compiler bug: minimal repro, version bisect, "we should pin the toolchain and file upstream."

It was a latent infinite loop in my own solver.

It's never the compiler. It's never the compiler. Scarily human-like tbh. 🙃

[Rasmus Ros @rosrasmus]

@brankopetric00 You probably don't have the problem that k8s solves.

[Rasmus Ros @rosrasmus]

@ibuildthecloud AI is surprisingly good at what i think is hard but fails what i think are obvious design decisions 🤔

[Rasmus Ros @rosrasmus]

@nummanali I'm in the same boat. I usually just set /goal and leave my PC then check back what happened

[Rasmus Ros @rosrasmus]

Good HN blog article on using LLMs at a slower pace: using multiple agents on the same PR to surface issues, then reviewing and filtering the comments yourself, is a pretty sensible way to trade speed for better code. https://t.co/rW1lLitUq1

[Rasmus Ros @rosrasmus]

@JackFriedson They always take the most obvious iterative step and never take the step back you need for good architecture.

[Rasmus Ros @rosrasmus]

@jonathan_wilke @cursor_ai That is a good idea. I use plan a lot, and one of the conflated issues is probably trying to make too big a change at once.

[Rasmus Ros @rosrasmus]

@asmah2107 Pretty much. If the answer can be checked quickly, the system gets better fast. If you need a human, a lab, or six months in the real world, progress slows down a lot.

[Rasmus Ros @rosrasmus]

@HoytEmerson Right. Don’t optimize for scale until you need it.

[Rasmus Ros @rosrasmus]

@cnakazawa It’s the other way around. If the Rust Bun has now is poor quality and full of unsafe, that’s not going to help adoption.

[Rasmus Ros @rosrasmus]

@haider1 Yeah im awake. What am i supposed to do again?

[Rasmus Ros @rosrasmus]

@8teAPi The conversation is driven by those who speak up. No way around it.

[Rasmus Ros @rosrasmus]

@tunguz Selling hand crafted artisanal C-functions $20. Custom bespoke assembly $50. Every line written on paper from local trees.

[Rasmus Ros @rosrasmus]

@sunnyyuych Efficiency in speed is one thing. When people get hooked on this technology it's too easy to shut your brain down and reach for it instead of searching or reading a book. I wonder tho if this is also an illusion or do you actually use less energy from just using your AI?

[Rasmus Ros @rosrasmus]

@JosephJacks_ The world is split in two atm: 1) people talking about AI and 2) people talking about people talking about AI.

[Rasmus Ros @rosrasmus]

@trikcode $20/month is cheap compared to hosting it yourself. And opencode is not as good as Claude code. It's good that we have the option but it's not a rational choice for individuals. If you're a company that changes the equation.