Madhusudhanan G

open source contributor
into Search of Re
make ways to llm and tmi

[Madhusudhanan G @gpkmasa]

My first research paper is now live on arXiv.
I looked at whether visible reasoning can help detect indirect prompt injection attacks across English, Tamil, and Tanglish.
A small study, with plenty left to explore but really happy to finally put this out.
https://t.co/nBGzRXCQ2y

[Madhusudhanan G @gpkmasa]

@tamilravi Well have to wait and watch

[Madhusudhanan G @gpkmasa]

I am not applying for @jio ipo

[Nick @nickcammarata]

interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had

with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works

https://t.co/zHij5zwKGT

[Madhusudhanan G @gpkmasa]

@fit_fr_nothing 😂

[Madhusudhanan G @gpkmasa]

@trusfratedpuzz Yeppp

[Madhusudhanan G @gpkmasa]

I got the thing why chatgpt is so popular among non techinical people

They are using it for personal things
Omg 😲
Literally I have seen lot lot of people

[Madhusudhanan G @gpkmasa]

@gunkejs Yesssssss

[Madhusudhanan G @gpkmasa]

@cvkrishnan Pavam 😂

[Madhusudhanan G @gpkmasa]

@TheAhmadOsman 😁

[Madhusudhanan G @gpkmasa]

@KarthikIO 😁

[Madhusudhanan G @gpkmasa]

@Tesla What is happening 😳

[Madhusudhanan G @gpkmasa]

@Vidhiyasb Not yet

[Madhusudhanan G @gpkmasa]

@VigneshAnguraj Great work 👏

[Madhusudhanan G @gpkmasa]

@ChaalPritam 😁

[Ramchand Kumaresan @Mechramc]

@realitycheckind @proxy_vector @MuraiLabsHQ Thank you! I started solo and then @gpkmasa and Karthik, joined as volunteers and helped with SFT data validation and testing!

[Ramchand Kumaresan @Mechramc]

It's a research preview: it gets facts wrong, so check before you rely on it. With your consent, we keep the conversations to evaluate and train the next version. Please don't type personal details. Open until Sunday 18 Oct, 19:30 IST. The paper covers how we built it: a tokenizer made for Tamil, 50B tokens of pretraining, fine-tuning, and what works and what doesn't. Model weights and code come after this preview. அகரம் — the first letter.

[Ramchand Kumaresan @Mechramc]

What helps us most: ask it real things you'd ask a friend, in Tamil, Tanglish or English. Try the 🔧 tools: weather, currency, maths, Wikipedia. Then tap 👍 or 👎. Every rating shows us where it's weak and we can refine it better based on what is not working.

[Madhusudhanan G @gpkmasa]

@Mechramc @MuraiLabsHQ Less Gooo

[Ramchand Kumaresan @Mechramc]

Agaram 2.5B, @MuraiLabsHQ's first Tamil–English model trained from scratch, is open for a 4-day public preview 🎉

Chat in Tamil, Tanglish or English, try the tools, and rate replies 👍/👎.

🔗 https://t.co/wmXQKAmBvm
📄 https://t.co/GBgleDor7U

Weights come after the preview - waiting on licensing to be cleared.