code for my public webpage: https://darshanmakwana412.github.io
9:53AM: Early this morning as I walking along the streets of HSR layout I was again visited by the surprising realization that everyone around me, each and every individual that I saw today had some desire in their life or still has that and they are actively working each and every day towards realizing it. Along the way the journey or the path that they have followed has shaped them and their…
Time and time I have found myself at different crossroads in life. These are the principles that I have developed by my constant failures, observations and interactions overtime to achieve unfair advantage and keep on winning
Since the last 2 months I had to change my environment in which I used to do my long hours of deep work. One of the things that was part of the change was a standing desk. Actually it was not a desk per se but just a platform on top of which I can keep my laptop and just start coding away, the platform was just tall enough that I can’t sit and do work and slightly short enought that I had to bend…
Gaussian Splatting is a fascinating scene reconstruction technique introduced by INRIA and last year I had a lot of fun tinkering with it while on my semex. I recently discovered some of my notes related to it and decided to digitize it this weekend, along the way I reimplemented the forward rasterization pass in rust and decided it would be fun to write a tutorial explaining gaussian splatting to…
We have spent the last few years training LLMs on human conversation data with the explicit goal of making them sound like us, but we have also been training ourselves on them in return. I have spent a lot of hours talking to claude over the past couple of months, my juniors grew up doing their assignments with gpt, and there are kids in undergrad right now whose first real intellectual sparring…
A couple of months back I was facing a kernel panic “Unable to mount root fs” while rebooting by ubuntu with normal mode, I found a quick workaround via recovery mode by dropping to root shell and patching some systemd files to boot with an older kernel version. This is how I had been restarting my computer since then. Today I saw by brightness keys where somehow not functioning at all, I dropped…
A lot of effort is spent to make LLM inference cheaper and performant. Quantization is the standard way to do this, where we reduce model’s size by representing it with parameters with fewer bits so they take up less memory and move faster through the memory hierarchy. The progression from 32-bit -> mixed precision -> 16-bit -> 8-bit -> 4-bit formats has been one of the most impactful practical…
One of the biggest hurdles I faced in consistently maintaining a blog like this site is having to manually copy paste my notes into my github.io directory as markdown files. This friction compounded over time and I would end up with a backlog of drafts that never made it to this site. So I decided to tinker around this a bit and create a more automated solution
I was listening to Dario Amodei’s interview with dwarkesh patel and found his insights into how anthropic plans their capex investments and path to profitability quite fascinating. They need to balance their risks into how much compute to build for the next 2 years in advance based on current demands because the data centers take 2 years to build. If they overestimate their demand then they won’t…
It takes very high agency To realize when you are in the wrong place To realize the world is drifting apart from the old rules and new rules are being written, the old rules which you learned no longer apply To realize the rewards tremendously outweigh the costs of consistently putting efforts on updating your mental models To realize that people don’t act with incentives as much as they think To…