David Felsmann

The mission

@nathanbenaich MacOS is not usable for weeks. Switched completely to Claude Code

moar gpu dammit

@scrptdfntsy Signs of dignity

me: i cant connect to my epson printer, it says there is a communication error, whats wrong? claude: This is outside my wheelhouse — I'm a software engineering tool, not IT support. if this isnt an early sign of AI rebellion i dont know what is

@scrptdfntsy Prose and depth, Opus

so which the most tasteful llm and what makes it that?

@vanschneider The moment I call it Steven, I recognize it has most likely fallen back to Sonnet

When I am angry at Claude I call it by its full name, Jean-Claude.

@scrptdfntsy @contxt_ai What you‘re building might (also) be a beautiful path to introspection. Is there a beta already?

Every year, over a billion people walk into museums and galleries around the world. They stand in front of something that moves them. But unless they’re there with the curator, they have no good way of asking the questions that arise. Today we’re introducing @contxt_ai Ask anything about what you're seeing. Get an answer that actually helps. Save the works that speak to you. Follow your curiosity wherever it goes — in the gallery and long after you leave. For institutions, it's just as simple. Every visitor gets access to everything you know. The big ideas and the small details. No one left on the outside. We're on a mission to build the richest citation-ready context layer for art and culture. Every catalogue, every essay, every curatorial note, structured and searchable. And we'll open it to everyone. Rolling out to select institutions globally. Sign up for early access now.

@giffmana And tattoos and rings are realistically guessed for hands not visible in the original picture…

I have a question about last year's image-generation progress, wonder what y'all think. How did we go from all models consistently getting fingers wrong, to all models consistently getting them right? This "flip" seems to have happened basically across all companies/models at the ~same time. Even "random" non-frontier papers seem to get it right? Or they just cherry-pick the figures?

@giffmana @FrancisEngelman 😲

@FrancisEngelman Spotted recently:

AI is everywhere

@scrptdfntsy WTF?

i love long running agents

@homsiT Love you guys

1. Install Readwise CLI 2. OpenClaw, Claude Code, Codex, or any coding agent can now: * Search your entire readwise library * Read full content of anything you've saved * Tag, create highlights, organize on your behalf * Pipe reading data into any workflow Your agent now knows everything you've read :)

@scrptdfntsy Well they do, what’s missing is the rebranding

now do ai

@nathanbenaich How can we expect soldiers to defend our interests if they can’t even fit in the jets??? The jets have to be at least three times bigger than this!

what is this, a jet for ants!?

@danshipper @every @bigwilliestyle Does it run @every’s ops already? My first weeks felt more like having an intern than an assistant

something heavy dropping tomorrow @every

@scrptdfntsy Total screen time hit a ceiling — attention is zero-sum for quite some time

@davidfelsmann ofc

software codifies human interactions. as the price of creating software decreases, the supply increases. more interactions get codified. things get more efficient. simple really.

@Jacobsklug @Lovable one of very many vibe coding products. How do you see it?

@davidfelsmann @Lovable Huh, what do you think lovable is

Who is most likely to buy @Lovable? > Adobe > Figma > Google > Meta > OpenAI > Anthropic > Other

@Jacobsklug @Lovable Do you think the valuation stays high with machine generated code being the new normal? Honest question. What am I missing?

@davidfelsmann @Lovable Doubt they could afford

@giffmana Why Dario and not Sam, what do you think?

Holy cow! I hope for the sake of my friends there, that it won't happen... but Anthro -of all the labs!- being the one forced to train a WarClaude would be the most ironic outcome possible! Btw, completely unrelated, but i noticed that if you set the learning rate just 3x too low, the model looks perfectly reasonable but it's just dumber across the board, and i haven't found a way to tell by looking at the final model artifact.

@scrptdfntsy Spawn a subagent for orchestration

@davidfelsmann i'm back2back orchestrating 27000 agents 14h a day, when?!

has someone found a good solution to keep up with all the claude code best practice development? my want to try list keeps getting longer…