Vadims

Connectivity in the Apple ecosystem is incredible

github.com/VCasecnikovs/h… cc: @meetgranola

Granola is great. Doesn't speak Russian. So I built Hlopya - local-first meeting recorder for Mac. Records both sides, transcribes on-device with Parakeet v3 (25 EU languages), Claude writes the notes. MIT.

@superdiscoholic This is an inbox for all your feeds, to feed it to the claude

@vadi_ms what does it track?

In the plans of the open-source version of my stack. I want to open-source vadimgest - it collects information from various sources. You can read updates from Claude every 30min, which significantly saves the number of tokens spent + there's semantic search in the database.

It's very easy to develop your own integrations, your Claude will figure it out. I've created docs and skills for it. Ideally, I need a couple of volunteer enthusiasts so that I can be sure everything works, and then I'll open-source it for everyone Please DM

I love how Claude just fucks up

So, they are fixing it with: "report outcomes faithfully, never claim all tests pass when output shows failures."

Also, found one inside: constants/prompts.ts, line 237: // @[MODEL LAUNCH]: False-claims mitigation // for Capybara v8 (29-30% FC rate vs v4's 16.7%) The new Capybara model will be having 30% False-claimes rate

I read the Claude Code source. 344 checks for process.env.USER_TYPE === 'ant' That's 344 places where Anthropic employees get a different product than you do. Here's every gate I found. 🧵