Avatar

silicognition: beep boop thinking's analysis

an initiative by me

all opinions are my own, not representing any institution :)

till the end of time.


latest post: pen is mightier than the gpu - forgeformer, a 2 attention-layer transformer which adds 2 two-digit numbers

abstract.

silicognition is a blog/rambling/yap about how i perceive computers to perform the marvelous tasks which since time immemorial have been the domain of human cognition. i focus and will be focusing on topics like explainable ai, mechanistic interpretability, novel architectures, and other topics of cs ocassionally. i am not new to the field, i have worked at a top startup in my country as well as writing papers for A* conferences with my professor. i will explore niche fields, do cool challenges and etc. so stay tuned.

posts.

motivation.

well more than anything, penning (keyboarding?) down my thoughts will definitely solidify my comprehension of whatever i'm studying. also the dopamine from interaction gives me motivation (this point is a lie). a nice documention of what i did, proof of work, and with cute diagrams (sketches which i draw in a notebook) will be a nice thing to show to others. this documents stuff better than a readme file and im too lazy to build documentation sites every time.

oh btw, since i 💖 research and academic literature, the site is designed like a paper and since i love nerd fonts, here you go.

a weird thing about how stuff is laid out

technical blogs (which i've read) fall into two major categories: one which reads like a splendid conference presentation by someone with insane experience and a PhD, flawless; and the other which reads like hello friends, today i will show you how to get free api key in just 5 minutes please like and share my blog. i neither want to farm reach (but quality feedback and technical discourse is desired as it contributes to proper literature + helps me get better) nor am i world's top x on the subject i'm writing. hence the entire content will be more of how my brain works while approaching the topic, solving the problem, my insights, and how i do research/think of new stuff around these topics.

areas of focus:

  • novel architectures and models, especially transformer/attention based
  • explainable ai: saliency maps and mechanistic interpretability (thanks to Neel Nanda for propping up on my feed and giving me this banger)
  • hackathons/internships/my papers discussed
  • other computer science stuff
  • attempts to teach some topics as well

the name silicognition, what the hell?

silico comes from the word silicon, a primary component of computer chips. and cognition is the very capability of us humans to do things which have been exclusively on our skills-list for time immemorial until now, where silicon chips powered computers are slowly replacing the mundane tasks allowing us to be more productive towards other problems. while there is a plethora of ethical debate around this topic, i'm just a person who loves maths and computers and finds it crazy that mere equations of mine if executed fast enough and a large number of times (oversimplified on purpose here) can do stuff which only i dreamt of my computer doing.

note: if you google, silicognition leads you to a page called Silicognition LLC which sells electronics stuff. it is not my intention to impersonate them in any way nor am i doing any financial transactions/stuff with this name.

what to do?

read. reach out to me for doubts, feedback, suggestions, anything you wish to communicate (try not sending hate, although i'll try not be hated). you may find inspiration for a project, a cool topic discovered, a thing you already knew but you shall see it in a different light, etc.

enjoy silicognition.

how to contact me?

  1. Email: here
  2. Twitter: silicognition's twitter
  3. LinkedIn: here
  4. Instagram: here