He fet aquest mapa que mostra Catalunya si les distàncies es mesuressin en minuts de tren a Barcelona (a Sants). 1 minut de tren és 1,2 km.
Si algú vol jugar una mica, us puc passar el link
Every week I receive some claude code stats and today I saw that last week cc worked for 103 hours. That's more than 14 hours a day. Still, I feel less productive than ever. I start 10 projects every week and finish 1. I can't keep my attention on a single task for more than 5 mins. Every time claude is working I move to another thing and forget the previous one. This week claude code wrote me 26k lines of code, but I can even remember 2 concrete things it did. It's like ideas feel less important than ever to me. I come up with an idea, start working on it with cc, and then, maybe after 1 single interaction, quit it. I can't imagine a worst brainrot level than this one, but sadly, I think we'll see it soon.
So, I started clipping long format videos we had at our company into short format videos (with captions and engaging hooks, using a tool designed for that). The first 10-15 videos I posted had 700-900 views, but now, all videos have 0 views.
I know those videos are not engaging enough, I consider them bad for tiktok (we're creating better shorts and will post them soon). But I'm afraid that tiktok algorithm has labeled my content as bad/not engaging and it's not gonna show any of my videos from now on.
Any ideas on what to do? I'm thinking about deleting the 0 views videos or even create a new account.
Just out of curiosity, where do you use claude code? I started with the terminal but then I tried the desktop app and I'm very happy with it. Do you see any advantages in any place over the other?
Instead of sending every tool's full schema upfront, claude code sends a list of tool names, then a runtime instruction telling the model: "if you want one of these, you have to call ToolSearch first to load its schema." The instruction lives inside a <system-reminder> tag injected into the conversation. Here's what I captured:
<system-reminder>
The following deferred tools are now available via ToolSearch. Their
schemas are NOT loaded — calling them directly will fail with
InputValidationError. Use ToolSearch with query "select:<name>[,<name>...]"
to load tool schemas before calling them:
AskUserQuestion
CronCreate
CronDelete
CronList
EnterPlanMode
EnterWorktree
ExitPlanMode
ExitWorktree
Monitor
NotebookEdit
PushNotification
RemoteTrigger
TaskOutput
TaskStop
TodoWrite
WebFetch
WebSearch
[+ ~130 MCP tools (Slack, Notion, Gmail...)]
</system-reminder> And same goes for skills, another <system-reminder> lists each one with a single-line description.
Those <system-reminder> are sent only inside the 1st user message of the conversation.
This architecture makes a lot of sense actually. In my case, the system instructions + reminders alone burned 38k tokens (I sent a “hi” message to test this). Loading every tool's full schema on top of that would be painful.