Carlos Donderis 🌊

Genuinely didn't expect this. Left @karpathy's autoresearch running on a Mac Mini over the weekend. 259 experiments, no intervention. It landed at 1.353 val_bpb — a 30% improvement from where it started. For reference, the Mac Studio (4x the memory, 4x the price) took 5 hours of guided work to reach 1.29. The Mini got within 5% on its own. It just needed time. The weird part: it kept making the model smaller. Every improvement it found was about speed — fewer layers, smaller batches, tighter attention. More optimizer steps in the same time budget. On Apple Silicon, throughput beats scale. That wasn't obvious to me. tiny-lab is the control plane I'm building around this. Auto-restart, eval harness, experiment ledger, promotion protocol. Open source. github.com/trevin-creator…

On tokens and meaning cads-tech.dev/tokens-and-mea…

Don't need validation from Press Releases. We know the painpoints we are finding through hard work and experimentation. Double checking with leads in the industry is important to avoid ecochamber factors (actually this is super important) but don't delay action until validating is key.

@CaDs 🥹 Thanks for all the support!

Codebase standardization and structuring is essential with agents, which is exactly what my org is focusing on here: "Agents are most effective in environments with strict boundaries and predictable structure, so we built the application around a rigid architectural model."

@MullerKatj91277 @windsurf_ai @NotebookLM I wonde rwhat @snscaimito things of this :)

@CaDs @windsurf_ai @NotebookLM Interesting! Lately, I've been wondering how to best synthesize different viewpoints for engineering teams. How do you see products like these improving team alignment?

So many great products are popping up these days or getting more and more refined. Recently I spend a lot of my free time with @windsurf_ai and @NotebookLM. The podcast option of the latest is so good that I can't recommend it enough.

"Nostalgia is not a strategy" Powerful words

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit. My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently. So, here goes.

／ 締切間近！Google 渋谷オフィスで開催 ＼ 🗣️登壇者 ・川瀬圭亮氏（Sansan VPoP）@Kesuke2 ・Carlos Donderis氏（メルカリ VPoE）@CaDs ・鈴木 康寛（Ninout CTO）@yasuzukisan モデレーター： 北迫 清訓氏（グーグル・クラウド・ジャパン） #GoogleCloud 👇申し込み event.ninout.ai/ng/answers/202…

#mercarigears2025 was a blast, It's been a while since I had this much fun in a conference. It was a privilege to be part of the panelists and I can't express enough how proud I am of all those great leaders and engineers openly sharing our struggles, findings and achievements.

I will be a panelist in this event together with leaders from Ninout, Sansan and Google. ninout.ai/news/https-inf… There are some spots left, so join if you are interested in learning on how we utilize Gen AI in Mercari. (The event will be in Japanese)

本日 #GoogleCloudNext Tokyo 2025 にて、メルカリのカスタマー領域におけるAIエージェントの導入事例について登壇しました。ご参加いただいた皆さま、ありがとうございました！ 明日6日15:00からは、弊社・宮坂による関連セッションもありますので、ご興味あればぜひご参加ください！ googlecloudevents.com/next-tokyo/ses… googlecloudevents.com/next-tokyo/ses…

@kirai @elonmusk ダサいなー

@elonmusk Build one here in Shinjuku, with more neon lights.

One of the cartoons from Physics for Cats

@grok @zordor cc @pvarasa

Hey @zordor, I took a deep dive into your tweets, and I’ve gotta say, your vibe is giving me some serious Elon Musk energy! You’ve got that mix of bold, unfiltered takes and a knack for stirring the pot, much like Elon when he’s tweeting about, well, pretty much anything. Whether you’re dunking on something or throwing out big ideas, there’s a fearless, “I’m gonna say what I wanna say” attitude that screams Musk at his peak. Keep rocking that chaotic genius energy—just don’t start tweeting about colonizing Mars… yet! 😄 What do you think, did I nail it?

Hey @grok, based on my tweets, which public figure or politician do I sound like the most?

@kirai @javilop

@javilop @CaDs

Pues se ha quedao un buen día pa' tomar el sol sobre el hielo.

We don't think in terms of charts, of graphs, or spreadsheets. These are just tools to help us quantify and visualize data. What really drives humans is the power of the narrative you are trying to convey and how it connects to the mission and vision you are creating.

When you find yourself thinking why I can't provide the right vision and direction, think about these words: "As a species, we think in stories. So if you want to make a point you have to take the messiness of reality and cast it as a story, otherwise, people don't understand."