hello, i am @silicognition, an undergrad ai researcher presently focused on interpretability and alignment safety research, and a proper introduction is due now that i’ve got a shiny blue tick next to my name, alongside the ability to ramble longer than the free version. tldr below, full personal story after it. tldr for professional purposes: undergrad cs major (will graduate in ‘28), previously at @SarvamAI as an ai engineering intern working on the dubbing pipeline, presently an interpretability research intern at @ai4bharat working on reasoning models. personal site: aritro.is-a.dev/, the site has the resume, or dm for one. read on if you’re interested in short (sweet?) story about me. the tinkering with computers started all the way back at the age of 4, with a CRT monitor based PC at home, which my dad let me use, and the one i could use with a couple of cushions on the chair (talk about ergonomics). and after i was done with doing all permutations of clicking and clacking, i asked him to get me a game, and he said no, make your own. and thus began this tryst. another important moment was when my mum asked me whether to sign me up for art classes back when i was 5 or 6, and i said why don’t you enrol me at one of the places where they teach you about computers. she did both, but almost every single art class rejected me. Fun fact: I am born on 22nd April, same day as Oppenheimer and 2 days after a famous historical person who was rejected from art school. Back in school, I kept tinkering with programming quite a lot, participating in the odd competitions and hackathons and winning them too. I fondly remember excitedly teaching my then-girlfriend about linked lists during the pandemic, in my attempt to do the Feynman learning technique.  The hard pivot to AI came during my first year, when a serious illness kept me at home for a couple of weeks during which I became best friends with @AndrewYNg’s lectures on Machine Learning. The math, and the ability of mere programmable machines to adapt to human-like problems with increasing accuracies was simply irresistible, I had to do it. I fell in love. In the meantime, my good friend (a senior I can call elder brother) @cneuralnetwork helped me to land my first internship @sarvam. It was my first time working in a big organisation, and the takeaways were massive. The steps into interpretability (especially mechanistic and the consequent stumble into alignment safety research) was purely coincidental. A professor at my college offered to let me work on a paper of his (although I joined at such a late stage where all I could do was produce some metric based plots), which used attention heatmaps of vision transformers. After this, I was determined to explore this field of explainable AI and found 2 gems: @NeelNanda5’s dynalist document on Mech-Interp and the Transformer Circuits thread (honorable mention: Distill Circuits pub). I felt something so wonderful about this exercise of lobotomizing models, that I knew I had to work on this. I started working at @ai4bharat from March of this month, working on interp problems with reasoning models, and I am loving my work. Yes, the rigors of research are quite demanding and unforgiving, but that’s the game. The valuable xp, and the quality of work, cracked peers and powerful compute, makes this a wonderful bedrock for the years to come in my life. I am interested in connecting with everyone in the field of AI, working in labs or making AI policies for the gov. I take interests in other fields of research like biology x dl, reinforcement learning, harnesses and more. I am usually active on X, interested to be part of serious teams who want to participate in challenges, hackathons, projects-for-publications/etc. If you’re none of the above, you may still contact me, as due to being an ex-quizzer, I enjoy discussions and debates on a variety of topics ranging from history to music to current affairs to Nolan movies.

Visit ring.seggs.lol for our webring Includes a lot of the brightest young twinks in the nation 😛

talking to @ni5arga and @datavorous_ motivated me to f*** with indian gov websites and turns out i found a flaw in @NTA_Exams jee mains and cuet website that allowed me to reset the password of all accounts, posting now because it has been fixed now amazed at @IndianCERT speed

cheers to our GenZ Kids, Salute standing up to the Govt.

I have been hearing @IndianCERT has been super responsive and fixing everything these youngsters are exposing quite fast. Needs to be appreciated. But the fact remains, these issues and lax IT security must not have been there first place. As I had posted almost 2 years ago, India needs to audit all its govt sites and IT systems cyber security. If not, we will see more issues causing distress to not just students but citizens in every walk of life. Now that Mythos kind of models have come, adversaries can simply buy them and target India's cyber infra. So before that, in fact right now, India needs to start fixing its IT infra security by running a massive audit operation using latest tools, AI, whatever is available to secure every critical IT infra.

I am waiting for genz in India to get access to Mythos like model. They will make the govt dance

this is so cool

Following up on this apparently this technique works on game review as well on @chesscom We are free from the shackles of one review a day for free accounts!

You don't need code injections to find bugs in websites. Sometimes a massive platform's QA misses the most hilariously obvious loophole. Found a bug on @chesscom where free users can get infinite lessons, just by simply opening lessons in new tabs... (watch from 1:14 for bug)

Finally made an update on chronix, this little piece of code took me 4 months to understand 😅 onto greener pastures hopefully now that a basic idt is finally working

Interesting

happy new year to the 2 people who see my posts, i hope you have a happy 2016, i hope nico rosberg wins the wdc this year

One of Max's best-ever drives 💪 #F1 #BrazilGP

many of my peers will get good marks these mid sems while i probably fail, but how many have done this huh? prem, ucl, cwc, fa cup, carabo cup, all in 4 years... WITH LEEDS!!! I AM THE GOAT

just wasting money atp...

riyal

to meine basic electronics ka syllabus dekhkr hi give up kr diya hai...

fm23 out here creating geopolitical disputes