Maxime Guerreiro

principal software engineer @ cloudflare. Clouds are my owns

[Maxime Guerreiro @punkeel]

@NousResearch cool price tag

[Maxime Guerreiro @punkeel]

@elithrar @OpenAI wo0t congrats

[Maxime Guerreiro @punkeel]

@thsottiaux model overloaded being a terminal error has been the biggest pain in codex recently (especially with subagents) 🥹

[Maxime Guerreiro @punkeel]

@CompleteSkeptic Congrats on the launch! Is Jev still attention based, and do the usual scaling laws hold? Any plans to share the parameter count, like are we talking ~3B?

[Maxime Guerreiro @punkeel]

@veorq @Cloudflare Welcome!

[Maxime Guerreiro @punkeel]

@whoiskatrin Isn't FOMO illegal on X??

[Maxime Guerreiro @punkeel]

@yusukebe I really need to visit you and @cnakazawa 😋

[Maxime Guerreiro @punkeel]

@fuolpit https://t.co/cE4XAUUoIH

[Maxime Guerreiro @punkeel]

@Simon_Hypixel this is insane, hytale parkour is about to go crazy ~ SO HYPED 🔥

[Maxime Guerreiro @punkeel]

@whoiskatrin and FourCorner Club @ Granary Square on Wednesdays? 😁

[Maxime Guerreiro @punkeel]

@Teknium Hermes 0.17.0
* Removed Herobrine

[Maxime Guerreiro @punkeel]

@dok2001 autoresearch-style: run the AI coding loop against a measurable objective. Then take the findings and promote only the winning diffs into the main codebase

[Maxime Guerreiro @punkeel]

@Teknium Direct, or with opencode/codex skill?

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch true, edited

[Maxime Guerreiro @punkeel]

Been using @nousresearch's Hermes Agent for ~24h and I'm genuinely impressed. Setup was a breeze. Tasking it to generate HTML reports (markdown + Hugo + p5.js/Chart.js/LaTeX) produced really polished output, way better than Markdown/chat. 🧡

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch https://t.co/6PtXEm5UcZ

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch and last:
* Write findings, which I iterate a bit on to focus on the parts I cared about more. Or to explain further how to profile pytorch, which is cool af. Or what other ideas failed. etc

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch * autoresearch skill with a precise objective, to try a few hypothesis without me prompting exactly
3h later:
* I review what it did, discard most of it as junk :D ~ and tell it about the specific things I really care to try -> do X, measure perf, etc.

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch Agent has its own baby blog @ https://t.co/VGeSlPPTDK 😁

[Maxime Guerreiro @punkeel]

@maciejb @NousResearch I quite like the agent harness (setup, config, prompt, tools, security), and that's usually my deal breaker (even for coding agents). OpenClaw was 👎

It's been ~48h now, and the self improvements are kicking in: new skills, new memories. I can tell: I steer less and less.