Nakul Arora

Build recommendation systems @KAYAK | Ranking, Search, Marketplace | Writing about AI, Product thinking, & what it means to build in the AI era | IIT → Tuck MBA

[Nakul Arora @nakularora]

@TheSpacerr All of the above. ;)

[Nakul Arora @nakularora]

People who think in systems will flourish a lot more than others in the new AI first world.. As AI reduces the cost of executing tasks, the real leverage will come from understanding incentives, feedback loops, bottlenecks, and how things connect. AI will amplify this leverage, and systems thinkers will benefit the most.

[Nakul Arora @nakularora]

@MartinGTobias Agreed. For me what works is having 1 top priority per week and everything else is good to have

[Nakul Arora @nakularora]

@dev_maims So does product management. It's changing so rapidly..

[Nakul Arora @nakularora]

@vitaliidodonov Absolutely, I did the same for my goals. Took a stab and dropped most of the stuff. I have also realized that designing my life around constraints is better as it helps drop irrelevant fluff faster than anything else.

[Nakul Arora @nakularora]

@zarazhangrui So true. Infact I find that using it early in my writing process biases the way I express my thoughts. I now only use it to brainstorm and think through the concept and avoid drafting with it.

[Nakul Arora @nakularora]

Building ranking and recommendation systems for a living has made me appreciate LLMs a lot more. IMO, referring to LLMs as "just next token predictors" is a gross oversimplification.

Anyone who has worked on prediction systems knows how hard it is to get predictions right. Predicting which hotel a traveler is most likely to book is hard. Predicting which result deserves to rank higher is hard. Predicting what someone means from a messy query is hard. And doing all this while getting results in front of the user with a super tight latency budget is insanely hard.

LLMs are a live embodiment of taking that idea and scaling it to language. Predicting the next word across code, research papers, financial reports, weather data, product docs, human conversations, and large parts of the internet is not a trivial task. For the model to do this well, it has to learn compressed representations of the patterns behind all of those systems.

IMO, that’s the part that often gets missed. The magic isn’t that LLMs “predict the next token.” It's that the prediction problem is so hard that useful reasoning, abstraction, and simulation emerged organically from solving it.

[Nakul Arora @nakularora]

@aiwithjainam This is why I love X. You keep finding these gems 💎

[Nakul Arora @nakularora]

Random thought from yesterday: presence might be the simplest performance unlock that nobody actually uses.

Not focus hacks. Not systems. Just being fully here for the thing in front of you.

Obvious in theory. Surprisingly rare in practice.

[Nakul Arora @nakularora]

@gdb ChatGPT is the best so far in terms of carrying context across conversations. It amazes me at times with the connections it makes esp when it remind me of my own priorities that I have stated elsewhere

[Nakul Arora @nakularora]

@reach_vb @petergyang Absolutely 💯

[Nakul Arora @nakularora]

@EMostaque @awxjack Some of my best memories are from my time in Dhaka working to help setup Teach for Bangladesh as their teacher training and curriculum lead. Bangladesh is a special place and it’s filled with people hustling to build a better life.

[Nakul Arora @nakularora]

@vitaliidodonov Agreed. The speed with which I work and the breadth of things I can handle have all improved multifold.

[Nakul Arora @nakularora]

“Worrying is the worst way to use your imagination” is one of my favorite quotes. I have it pinned to my wall and every single time I read it, I end up laughing because more often than not, I was in the middle of a deep worrying session.

[Nakul Arora @nakularora]

So true and it’s such a critical skill to develop. Writing clearly helps you get better at articulating and storytelling. It also forces you to know your stuff inside else the lack of substance is easy to spot.

[Paul Graham @paulg]

The most important component of writing clearly is simply to have high standards for clarity. Then if you write something unclear, you notice, and ask: what did I mean to say? You can just keep doing this over and over. And if you have high standards for clarity, you will.

[Nakul Arora @nakularora]

@iAmHenryMascot @pejmanjohn So cool. Thank you for sharing

[Nakul Arora @nakularora]

@sunchoi_go @TTrimoreau Lol 😂

[Nakul Arora @nakularora]

@sflorimm I use and love them both. I have found Claude works really well for planning and for execution I prefer codex. What about you?

[Nakul Arora @nakularora]

One of my biggest learnings in life is that if your ability to feel peace is tied to completion, you’ll never access it because growth is an ever moving target and never ends.

[Nakul Arora @nakularora]

A major side-effect of the AI wave has been the exponential increase in the size of everything. Be it a strategy doc, roadmap, jira tickets or even presentations. Everything has a similar flow and uses a lot more fluff than needed. Anyone else seeing this too?