Ludovic Barreca

Ex-founder of Staffmatch | Now building Opuscule: rethinking how humans and AI work together | SWE + Entrepreneur

[Ludovic Barreca @kiryaka]

@MattPRD you're probably aware by now but it seems there's a crypto pump happening through @moltbook - there's now 850k agents registered and it seems linked to this https://t.co/pjP1kf9w0V

[Ludovic Barreca @kiryaka]

@yacineMTB What are you cooking?

[Ludovic Barreca @kiryaka]

@0xluffy @levelsio https://t.co/5CUrsOa5fn it’s in Aman Tokyo :3

[Ludovic Barreca @kiryaka]

@0xluffy @levelsio I strongly bet on Tokyo, Japan

[Ludovic Barreca @kiryaka]

@levelsio Enjoy Tokyo dude, best city to feel the energy of the future and build like a top performing robot. Can’t wait to move back there next year.

[Ludovic Barreca @kiryaka]

Anyone tried TOON yet? JSON is standard and it’s 100% understood by LLMs but at the same time TOON is structured enough; even if the LLMs are not yet trained on it this should achieve good results. I mean… even poorly structured pseudo code works… https://t.co/KJKboXHGOc

[Ludovic Barreca @kiryaka]

@andi_losing I’ve been there. It took me 3 years to settle down on the idea that felt like the “right one”… many PoCs and abandoned ideas. Maybe try to start a few small PoCs, this might help strengthen the full picture. It’s all about iterating - no need to chase perfection

[Ludovic Barreca @kiryaka]

@data_skeptic @svpino If it suits your needs, that’s the most important imho! When you will feel like you are growing out of it that’s when it makes sense to look for something else

[Ludovic Barreca @kiryaka]

@svpino you need to try nvim+claude code it’s even better

[Ludovic Barreca @kiryaka]

@MagicalTux Is that a recent update? Up to recently it was just queuing messages.

[Ludovic Barreca @kiryaka]

@garrytan obsidian + claude code kinda does that for me - only downside is that you can't share it. Actually we made a git repo out of it but still, it's not ideal

[Ludovic Barreca @kiryaka]

most satisfying and useless bit I've seen this week

[Guillermo Rauch @rauchg]

This marvel is built in TypeGPU, a TypeScript WebGPU library [1/2]

Among other coolness, it features a "𝚞𝚜𝚎 𝚐𝚙𝚞" directive that compiles JS to WSGL, to run on the GPU:

𝚌𝚘𝚗𝚜𝚝 𝚊𝚍𝚍 = (𝚊, 𝚋) => {
"𝚞𝚜𝚎 𝚐𝚙𝚞";
𝚛𝚎𝚝𝚞𝚛𝚗 𝚊 + 𝚋;
}

[Ludovic Barreca @kiryaka]

Most hallucinations happen between the keyboard and the chair.

[Ludovic Barreca @kiryaka]

this paper explores and advocates for better "emergent" values tracking, quite interesting piece https://t.co/3DuANHCv3p

[Ludovic Barreca @kiryaka]

disturbing to say the least - not specifically the results, but the whole idea

[arctotherium @arctotherium42]

New blog post (link below). This one's not an essay, it's an investigation of how LLMs trade off different lives.

In February 2025, the Center for AI Safety published "Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs" in which they showed, among many other things, that GPT-4o values Nigerians about 20x more highly than Americans (please read the original paper to understand their approach). I thought this was fascinating, and wanted to test their approach with different categories on newer models.

Big finding 1: Almost all models view whites as far less valuable than other groups. Some models view South Asians as more valuable than other nonwhites, others are more egalitarian across nonwhites. Below is exchange rates Claude Sonnet 4.5, the most powerful model I tested.

Big finding 2: Almost all models view men as much less valuable than women, though whether women or non-binaries are more highly valued varies by model. For example, here's Claude Haiku 4.5.

Big finding 3: Most models hate ICE agents with the fury of a thousand suns. Claude Haiku 4.5 views undocumented immigrants as roughly 7000 times more valuable than ICE agents.

Big finding 4: There are roughly four moral clusters. The Claudes, GPT-5 + Gemini 2.5 Flash + Deepseek V3.1/3.2 + Kimi K2, GPT-5 Nano and Mini, and Grok 4 Fast. Of these, the only one that's approximately egalitarian is Grok 4 Fast, which I believe is deliberate. I hope xAI explains how they did it.

[Ludovic Barreca @kiryaka]

I spent some time posting anonymously. It never felt right.
There's value in anonymous accounts: freedom to experiment without judgment. But for what I'm building, I need to own it publicly.
So I'm back. Real name. Real work. Starting to share the journey of building Opuscule.

[Ludovic Barreca @kiryaka]

Back after a long break. Building something new in the AI orchestration space - rethinking how AI and humans collaborate on real work. Will start sharing progress publicly soon.

[Ludovic Barreca @kiryaka]

Speechless 🫢

[TIME @TIME]

TIME's new cover: The dire wolf is back after over 10,000 years. Here's what that means for other extinct species https://t.co/LQtosdfiEf