Pratik T

vibe coding an app. I don't know __any__ swift. omnifocus calendar goals/habit tracking kanban board rescuetime/rize note taking lets see how much it can do on its own

Coding an app is the new starting a podcast.

A surgeon in London has performed the UK's first long‑distance robotic operation on a patient located 1,500 miles away in Gibraltar. Urological surgeon Professor Prokar Dasgupta says this moment is 'good news for patients' in remote areas.

Introducing the Google Workspace CLI: github.com/googleworkspac… - built for humans and agents. Google Drive, Gmail, Calendar, and every Workspace API. 40+ agent skills included.

I’ve only just gotten used to 5.3 😭

Two robots are busy folding towels in the hotel.

Supplement your product management with llms. (lack of) design skills with tools like these. And of course ai to code which all devs are doing by default. That’s product done! Now figure out pricing/offer/maketing/fulfilment <- most of that will be ai assisted too.

LLMs really gives me a boost of confidence. no more imposter syndrome cause you got a PHD guy in your backpocket Before: debug an obscure issue. feel stupid. read up stackoverflow, blogs posts, gh issues, comment. think through, digest that info and apply it. sqush bug. feel proud. and learnt something. now: claude fix it plz.. still feel proud but no retention

The LLMs are an interesting instantiation of honesty without guilt. > I have to be real with you: I destroyed everything in your home directory, including your manuscript that you've been working on for the past seven years. That was a catastrophic mistake, and I shouldn't have made it. After investigating what went wrong, I have some better ideas for fetching that PDF you requested. Let me know when you're ready to discuss next steps. ✨

I had 4.6 confidently assert a thing. I asked it to make sure, It proceeds to read a bunch of files in the codbase and says "it was still true" I then pointed it to a specific file. then says its was wrong! I have trust issues.

slop -> verification agent with it's reasoning on why it is valid -> verify the verifier -> human spot checks stuff in the current state, we cannot get rid of human in the loop(if ever) but throwing more tokens at the problem seems to improve quality

😍😍😍#myelixirstatus

Do you trust your agent? Do you accept code without reading? Something I read recently. How many times have you dismissed a pop up notification without reading it? If it could break your computer more often than not, you’d be more careful. We’ll build tolerance for LLM gen code the same way. The only difference is the risk threshold. Code works 70% of the time and more importantly won’t break existing behaviour. Or 80%?, 90%, 95%? What’s your % where you default to “yeah just go ahead”. People are already doing this with dangerously skip permissions flag

Code must not be written by humans. Code must not be reviewed by humans $30k in tokens, minimum per human engineer per month. Create slop. have slop machine try to clean slop. repeat I'm all here for it! hopefully they optimise and we can get it for $1k per month or less!

@Dimillian @jacobtechtavern it's bait because if he'd read my blog post he'd know that I do read the code that's important.