Wow, incredible

Just saw your other post: Moonshot AI's OK Computer. Very cool.

One hour straight without any other human input? What's the name of the tool? Thanks for sharing!

Can you share an example? I think my prompts are around 20 words typically, although I’ll often ask it to construct a much larger spec document and I’ll include a bunch of relevant files so the total context is much larger.

Hi can you share the source for this? I just see the options to connect Drive, Gmail, and Google Calendar in my app.

Hurts all the way! Also, he helped build the stacked roster. He made sure his contract left enough salary cap space for his teammates. He's on another level.

Whenever I try to install a DXT I get an error saying I need Claude Desktop 1.0. I’m on v0.11.6 but there’s no option to update. Any idea what you did to get v1.0?

Okay, this is after a single prompt to Claude with AI on: "Make a working text adventure starship bridge simulator. there should be different stations to control, and each should use AI in different ways to advance an overall story using Claudeception"

claude.ai/public/artif...

AI capabilities are doubling every six months. It’s okay to think AI is sucks today as long as you also have an open mind and re-evaluate constantly.

AI is really bad at some basic shit, but also incredible at some pretty difficult things. In that way it’s not so different from any other tool.

Mind providing an example? Sounds like I would provide a bunch of time series like monthly active users and monthly revenue and it can forecast them?

I've only played around a little but so far I prefer tools that support MCP (i.e., Cursor + Claude Code) so they can integrate with other services like Jira and Linear.

On the other hand, Codex is good for executing tasks in parallel.

I might be wrong but I think the effort is still worth it because you’ll have learned the skills to overcome the next model’s limitations.

Am I understanding this correctly: Dolphin makes it so OpenAI models can use existing MCP servers as well?

*just* getting started with Cursor + MCP. We wrote some custom tools to get Github PR review comments and Rollbar error stack traces. The idea is to give cursor a Rollbar error, have it submit a PR to fix the bug, and also address requests for changes.

Didn’t know canvas could run code! So it’s like Claude Artifacts?

Wow, an LLMS.txt file is such a simple but great idea

Dumb q: Why are there multiple huggingface options for the same model but with vastly different download sizes? What's the catch with the 16GB option? Thanks in advance!