Man, these GEPA instructions can get lengthy.
Who's working on 'regularization' for prompt optimizers?
"Shorter, lower-perplexity optimized instructions generalize better" is the new Occam's razor.
That feeling when your girlfriend from high school finally gets engaged.
I used to tell myself: Andrew, it’ll be ok. She wasn’t right for you.
I just hope her husband isn’t some handsome buff football player.
Started taking creatine monohydrate (5-10mg) per day recently. The biggest surprise is that I feel way better on less sleep. Used to feel like a zombie with < 6 hours of sleep - now I'm fine. This is after just 1 week.
How I set it up:
- Create a Slack app on https://t.co/Fn8uhcz34B and add scopes for a user OAuth token
- Run a Slack MCP server (I tried https://t.co/5FEnRj5fdM, but this one had a broader tool set and worked better for me: https://t.co/h7lFhWxOst).
Add env vars for User OAuth Token and Slack Team ID.
One tip that made it significantly more useful:
Agents often try to enumerate every DM/thread/channel when you say “check DMs for anything urgent.”
A simple workaround: tell Claude to use the search tool for conversations `to:@samad` (since yesterday). Keeps it focused.
Integrated Slack MCP with Claude Code last week.
Helped me a achieve a long-time side quest: Going (almost) one full day without opening Slack.
I can now:
1. Get a Summary of my `@mentions`.
2. Check DMs for anything urgent
3. Send out quick replies on threads
without leaving my terminal
- "i have no technical ability"
- "then what are you being paid for?"
- "the confidence i have in my taste and the ability to express what i feel"
the average vibe-code founder in 2025.
Have been using iPhone mirroring for a while now and it has helped solve a key dilemma I have been facing for a years:
- Distracting apps (like WhatsApp) logged in on laptop -> Easy to get pulled in
- Logged out & phone away from me -> Keep getting up for sending a small text, 2FA, etc.
- Logged out & phone near me -> Still distracting
With mirroring, I get the essentials without the pull to pick up my phone. Surprised more people don’t use it.
Overall, this feels like an early prototype of the future where autonomous agents quietly getting your work done, making smart suggestions and taking decisions on your behalf. Next I'm going to try audio input and see if I can do some "vibe-browsing"
Playing around with the @PerplexityComet browser today for a task I struggle with every week - buying a week's supplies from the Indian grocery store. Since Anthropic’s first computer-use model, I’ve been testing these AI-driven browsing tools every few months. Let's see how it goes.
Other interesting observations:
1. ChatGPT chats show the title instead of https://t.co/JFniXZWsuY. I found this delightful.
2. When I type "chat" , the first result in autocomplete is "https://t.co/tGvAFsrlKh" (which I haven't used in months) instead of "https://t.co/FHwZPVvDeB" which is the first result in my chrome profile? Would be interesting if this were intentional