I just started a community to share these: x.com/i/communities/…

Hey OpenAI , your iOS transcription function/service in the ChatGPT app is possibly the most unreliable service I have ever used. Congrats.

On Monday I captured 225,000 photos of the moon to create this 500 gigabyte composite showing more detail and color depth than usually possible through a telescope. Here's a thread about how it was captured, but make sure to load in 4k and zoom in first!

Here's an AI-pessimistic prediction- If AI becomes the intermediary interface between ourselves and all Internet content, then it will be ever less useful to produce said content anymore (at least towards a human audience) and there will be a rise of API-only interfaces to

I made a thing to make working directly with binary data in text-only contexts cooler (albeit at the cost of 0% best-case to 200% worst-case more bytes). LuaJIT and C implementations provided. Advantages over hex (xxd) and base64 (and as I discovered while building this, `bat

Functional and typed languages will end up benefiting hugely from LLM-accelerated coding. Why?

Hey Anthropic , Zed and @windsurf, I think your live/agentic code editors would benefit from keeping some portion of the current project directory file structure visible/in-context. Perhaps filtered somehow or depth-limited etc. Something like what the output of

I've been working on an automated AI debate tool called "tensor-kombat" (currently only a TUI app for now- Idris Language with the C backend, actually!) to experiment with using a restrictive functional language with LLM code assistance. Here's an example of its most recent output.

Dear SoundCloud , You seem to take much longer to open on iOS lately. What’s up?

🚨 JUST IN. Anthropic gave Claude $1000 to run a shop. It lost money every single day. But that's not the crazy part. It rejected 566% profit margins and gave away inventory while claiming to wear business clothes. If you think AI will replace workers, you need to see this:

Hey OpenAI , Grok , and Anthropic : I just had a possibly brilliant and startlingly simple idea: “3-player AI Chat” Basically, an LLM conversation that two real people can participate in at the same time, where the role of the AI could be mediator, factchecker,

If he does leave this up, no credit will be given to him

gist.github.com/pmarreck/cf199… I should probably promote this to a proper github project (it's currently just part of my dotfiles) but this is `rm-safe`, an implementation I made of a "safer rm" inspired by Solaris' take on this problem.

Hey Zed , Just FYI, getting this error currently when attempting to use Qwen3-Coder on OpenRouter:

I 100% vibecoded an app as an experiment in 100% vibecoding- an Asteroids clone with some “extra sauce”… Works on mobile and desktop browsers, and is entirely contained in a single html file, including a unit test suite that can also be run via Node in the terminal. Still need

The fact that Python is the standard for machine learning is a serious indictment of the field’s engineering standards

#NewProfilePic