Ishan Ramrakhiani

@rithvanga talk about premonition lolll

Cursor is losing because they can't RL foundational SOTA reasoning models for their harness. Only model providers can now win the AI coding wars. Chinese model provider distillation isn't sustainable and will always be behind the frontier.

Thank god I skipped Delve and just had Claude generate a SOC-2 report directly.

@j0hnparkhill do u have a good benchmark for us to surpass

-- Hiring ML Engineers time! -- Are you a smart person who wants to make therapeutics to help people using your ML chops? We will hire two engineers. We don't care about your pedigree, just your skills. Prove them by attempting the exercise linked below (<6 hours) 🧵

@arpitingle whats ur bpb

time to cook

@itsandrewgao @willdepue @philipxmeng comp

nooo how am i supposed to parameter golf when there are no 8xh100s help @willdepue

@garrytan @conductor_build have been doing this from the start. it's important to get a vibe check on whatever nonsense is going on so i can stop it quickly

This is amazing, I have my own experimental mode now in @conductor_build: Expand all tool calls One of the weirder things I realize most people don't do that I do: I read all the thinking traces of Claude Code and it helps me a/ understand what is happening and b/ stop Claude from going down the wrong path. I recommend you try it.

@0xrandomlabs it got a lot of its background info from @nozomioai

I built an agent harness based on @0xrandomlabs' Slate for long-running AI/ML research. Gave it a drug interaction prediction problem before English class. Came back 6 hours later to a SOTA model. Github: github.com/himanalot/DRIFT

Tried conductor for the first time since it launched, and I'm amazed. I've gotta say, @garrytan is right. Now that agents are good enough to use from this type of interface, without as much monitoring, moving fast is even easier than it was from my terminal.

@0xrandomlabs thx! ur blog was great

@fioraapp Insane work!!!

@0xrandomlabs Yes! It is.

@fioraapp Is this real

I built a long-horizon agent harness for research and reached sota performance on drug–drug interaction prediction models in 6 hrs. x.com/ChemVagabond/s…

@garrytan I think this isn't just limited to code. It works for the logic overarching the code as well. I fed the blog post for Slate into claude code and was able to scaffold out a system for designing ai/ml research experiments pretty quickly.

One of the interesting things you can do easily now is once you know a given codebase does X, any codebase you need to X can do it in about half an hour Suddenly hoarding code does seem like a great way to be able to do more things And more begets more

I realized something else AI has changed about coding: you don't get stuck anymore. Programming used to be punctuated by episodes of extreme frustration, when a tricky bug ground things to a halt. That doesn't happen anymore.

I built this out w/ video understanding, context management via a graph, RAG over scripting docs, etc. I learned that the current editors aren’t meant for agents to use easily. Outputs get better w/ new models, but this still stands. Spatial understanding & cost are also issues.

@ericzakariasson insane bro

yc companies now get $60k+ to try cursor across the team! 6 mo teams + bugbot + $50K usage credits you can redeem on bookface bookface → deals, or email yc@cursor.com mailto:yc@cursor.com if you can’t find it excited to see what y’all will build!

github contribution graph for imessage (ignore the message 😭) thanks for the inspiration, @sabziz

Sleeping 2h earlier significantly improved cognition & mental health.

@RishabJainK agree, but it's still rly useful for other types of apps

first time using flask over express in a number of years and i will say, never use flask it is completely incompatible with the web in 2025