VR

Well 🤔

This is exhausting!!

@marclou Only doing 2nd and 4th, will try all of this for a week

The morning routine: - Phone turned OFF last night - Breakfast w/ wife - 1–2 hours at the gym - Cursor + localhost:3000 only, until lunch That’s the only way I’m productive and happy.

Wondering if i am procrastinating more or less after i started using LLMs #claude #chatgpt @grok

Boring is better

Maths is fun… !! 🤩

3/3Also shipped: — Guided JSON (constrained decoding) → parse failures near zero — Prefix caching → faster TTFT on repeated queries — Persistent HTTP connection pooling → cuts loopback TCP overhead

2/3Qwen3 has a "thinking mode" — silently injects <think> blocks before JSON output, breaking all parsing with zero error indication. Disabling it needs TWO flags: --reasoning-parser qwen3 --default-chat-template-kwargs '{"enable_thinking": false}' Miss one → thinking re-enables

1/3Migrated a prod NL-to-SQL system from llama.cpp → vLLM yesterday. The migration took 2 hours. One Qwen3 gotcha took longer.

@claudeai Absolutely Brilliant…!! Crazy good…!!

My feed is full with launches everywhere, but most of those are just doing same things again and again. Like for example: Everybody seems to be building cross platform posts app… I mean why?? Is there something that I don’t know?? Happy to learn…!!

Are ppl still building AI chatbots??

Back to basics

Now then, Tenants can log in, create an entity, and see it appear in the dashboard. Dummy works!! Took a short break for Holidays, now back to the process.

WTF is wrong with cursor!! Even setting cursor rules is also not good enough.

Auth flow is live! Supabase JWTs validating perfectly and tenants are isolated in Postgres schemas. Nothing fancy to look at yet, but it’s the kind of invisible progress that keeps future headaches away.

@AdityaShips Share this with 3 people 😄

Course? Wtf. And $997 is unreal.

Sam Altman teases GPT-6 will not be about raw power, but memory. The next-gen model will remember your preferences, debugging style, tone, recurring bugs scenarios, long-term projects to feel like a real personal assistant. #GPT6 #AI