Ayush Goyal

If you are someone who is into post-training models over unsloth, we just made your life 10x easier.

@henrytdowling currently its reinforcement learning, down the line its going to be either RL sample efficiency or interpretability

@goyalayus What kinds of research do you want to do?

I am going to become one of the greatest researchers, pair up with a great systems programmer and a great designer and take humanity to the next level

the algorithm is too much slop

@geometricZed I have not seem much of it used

@goyalayus If u go for linear algebra, try Linear Algebra Done Right by Sheldon Axler

I am stuck on this, I have tried. tried a lot. every llm. none can make me understand it. i'll have to use some brain, i hate it. I just have to read it. no immidiate use. no beauty.

@geometricZed i wonder why linear algebra is useful for deep learning.

@geometricZed its a good book sir

@goyalayus Man wtf , u won't believe me but that's the exact book i am following 💀💀Co incidence?? Damnnn

@geometricZed probability by blitzstein

@goyalayus What book is it from ???

So, computer science is a game whose rules are created by humans, and then all the algorithms and all the optimizations just respect those rules. Mathematics is a game where the rules have been created by the universe, and we don't know the rules. So, we figure out algorithms and test if they match the previous algorithms. So, whenever you find an algorithm in math, it becomes a rule, and the next algorithm has to respect the previous rule, the previous algorithm. So, I think models will become really great at optimizing or discovering every algorithm for rules when all the rules are known. But when the rules are unknown, like in mathematics, models may or may not saturate it. But rules which have been made by humans, models will saturate them. For example, systems programming. But the thing is, the rules of physics are not made by models, so the rules of hardware are kind of not made by models. And then, so when the rules of hardware change—no, so because the rules of hardware are not known by models or humans, the algorithms can be discovered iteratively, and that makes the roots of software update. So, I think that's the reason models will not be able to saturate computer programming, also.

some weeks back I made a thing github.com/0xV0YD/Mnemosy… because I kept losing the thread between agent chats it stores the conversations along the way, organizes them into useful memory, and uses retrieval algorithms to remember decisions we made earlier when a similar topic comes up again. not limited to Codex either: plug it into multiple IDEs/editors through MCP and the memory layer can travel with the work. local-first, searchable, source-cited memory for coding agents. u can try it out

I am kinda happy I took the JEE; I think it really helped me think. they could have, though, removed chemistry from it and focused on going deep in a few chapters in phy and maths, and just covering other chapters like complex numbers in shallow depth

@0xVoyd proud of you son

so the difference is that DP already knows what moves it can make then it figures out corret possible ways while RL also have to fiugure out the possible moves along with the path as the possible moves will be the ones which would lead to the correct goal

@notthatkush sorry

people in 2026: 1) without reading pass the problem to claude/gpt 2) don't think at all 3) solution's ready it is crazy how often i see this during interviews and when meeting people who seem really smart. it is honestly bonkers how many people have genuinely forgotten how to think

Can somebody launch a DeepSeek 4 Flash-level model? Not V Flash, V4 Pro-level model, for ultra, ultra cheap? Like, if I use even a billion tokens, it would cost me a dollar or something. That would be great.

I was reading some stuff from... There's this guy who is an expert on async RL, and I was reading some stuff from him on RL algorithms, and I just realized that, fuck, I don't know any crop stats. I just forgot everything I knew. So, very ashamed of myself. I want to be a researcher, but then I don't know the basics. So I went in and thought that this would be the last time I would be studying mathematics. But this time, I would be studying it for life. By life, I mean I will not forget the things that I am studying for life.

@goyalayus Kya hua bhai itna frustrated kyu hai Maths ko leke ? 😭😭

I feel sad for the people who took crash course in mathmatics

If you are studying mathematics purely from Gemini, I would not recommend it. Pick up a good book. For example, if you are... Yeah, so pick up a good book and then study from that. Use Gemini for considering your signs. Also, Gemini Flash sometimes hallucinates. It will give you proofs that are not useful. So use thinking mode sometimes when you need it. It's useful to regularly think.

@goyalayus Us , I've been studying Maths with the assistance of Gemini flash extended nd it just feels way superior to GPT (have used gpt top for same purpose them switched to be in nd I'm glad I did)

I constantly find Gemini 3.5 flash a great teacher, somehow it gets me.

@youdesignn40682 how sure are you at historical humans were only based on survival majorly and not on love

@goyalayus They are something thats constant .. but if we talk about Historical human .. they were based on survival , love was idea of elite

I wonder, someone born in 18th CE, what meaning do they have to their life? They can't fly in airplanes. They can't live in air conditioning. They can't have many friends because if they move out of the city, they lose them forever. Their spouses, they can't choose their spouses either. They have to work all over their life. Life is pretty miserable, you know. I wonder why they would not be depressed. But then I think about myself, and some other Ayush who is from two years in the future would be looking down on me and saying the same things about me: this motherfucker has to take a fucking flight to go somewhere. Can't they just use anywhere doors? What's the meaning to their lives? They can't know the future. How do they make decisions? Got to know the future, how do they make decisions, etc.

@youdesignn40682 what is someones purpose is love and happiness, what timelines

@goyalayus Purpose is always relative to your timeline anyway ig

I'll try to finish this in one go but lets see