Read it again and again, a true compass for anyone paulgraham.com/greatwork.htm…

everyday, there is a new term on agents, but i see complex adaptive systems are the best first principle on building with agents and building agents! pracha.me/artifacts/posts/ag…

Started to build agents, then realized why not build agent that builds domain specific agents! Calling it as Von Neumann machine 🥲

How would you measure the revenue/sales brought by Advertising on trams 🚃 of Bay Rea Vs subways 🚇 of NYC ? and, if we could use synthetic humans to simulate these scenarios - how do you go about it, Like managing the confounders?

I believe Qwen is going to be the most trained and adapted model federated across enterprises from the agent traces! Like a shift left: start building agents on the costly tier, get the traces, distill, and train on Qwen!

Training any models?

checkout substrate - a environment i am developing to replicate enterprise data, like how data and context spread across different systems( time series, structured, semi&unstructured) , so that any agents for use-cases like analysis, inference, prediction on enterprise data can be developed (essentially like pre-trained) and further optimized. I have just started this as a side project for another idea i am building. might be useful if you are exploring agents in the enterprises setup! github.com/prabakaranc98/sub…

This is more illustrative and interesting

Before I start my summer internship tomorrow, I've captured a decade of learning — how it scaled over the years, why I've backed a curiosity- and exploration-driven path, and the few principles I've carried the whole way. Now the mission turns outward — from scaling how I learn to scaling what I build, research, and the impact I can make. And this time in a much more accelerated way: the learning is close to real convergence, and I intend to make it faster, bigger, better, and bolder

Remote sensing the great lakes

Just wondering, when all the reels and shorts from Google or any FAANG employees only say “do LeetCode, system design, competitive programming,” why do we need so many redundant interviews? There is no new perspective, nothing curious about the art of building software systems. Explore, guys! Build your taste—not only for research.

Last fall, I wrote that I wished to move to the Bay Area for my summer internship. So I am moving to SJ and will be working on AI and Digital Twins for autonomous network management and security. Apart from the internship, I hope to meet amazing people and find problem statements to work on. Over the quarters, I realized through conversations with founders and leadership teams that I can do and help much better in terms of AI/ML for many startups and enterprises. Let's just build things!

Tbh, I am a good fit for ai pm and ai fde lead 🫴

My mom doesn’t know exactly what I’m studying; she just knows I’m doing an MS. While explaining Data Science and AI—how we make machines think and act like humans—I mentioned self-driving cars in SF and that I might commute to the office in one every day. She was surprised. In a spontaneous conversation, I told her I might work on designing kids, ending cancer, forecasting the world, sending humans to space, growing lab meat, and more—whatever is possible with data and AI. It felt like a manifestation, as manifestations are spontaneous, not deliberately crafted. I did the same around 2020 when I said I’d go to or settle in Seattle, and my first MS admit was from UW. I’m not here for short-term indulgence. I’m here for the long game.

Give it a read complexityexplained.github.i…

X is the RL env for humans for hill climbing!

Yes, that's how it should be done. Ask about expectations in terms of salary, learning, and experience (and vice versa). Agree if it's good, then proceed with a better interviewing method.

IMHO, the salary has to be based on experience, skills, and the role, not the mere percentage increment or 🍎 vs 🍊 comparison of companies. When I left my first job at 14 LPA, the next role's max I negotiated was 32 LPA (with 3 YOE), and many were able to accept the number as I had experience and justification—so is everyone. Few offered max 25% on current CTC, and I didn’t even proceed with those interviews. So, it's better to set and exchange expectations in the first HR round itself, to avoid wasting time and energy on further interviews and potential drama at the end.

Would there be a new employee review system trend that somehow evaluates frugality and proficient aspects of an employee against token consumption?