Six Researchers Who Shaped How I Think About AI

I did not grow up around AI labs OR Computers. But When I finally found my way into machine learning, it was not through a classroom. It was through the writing, code, and talks of a handful of people who made the field feel reachable instead of sealed off. This is not a ranking. This is a list of my top Ai researchers.

Andrej Karpathy

Karpathy is the reason I believe you can understand a neural network from the inside instead of just calling an API and hoping. His CS231n lectures and his "Zero to Hero" series do something rare: they build things from scratch, in public, with no shortcuts hidden behind a library import. Watching him write a tokenizer or a tiny GPT by hand taught me more than any textbook chapter or College on backpropagation.

He has moved through some of the most consequential rooms in AI, OpenAI as a founding member, Tesla Autopilot, back to OpenAI, then his own venture in AI-native education, and more recently into pretraining research work with Anthropic. But the through-line has never changed. He teaches like someone who remembers what it felt like not to understand, and refuses to let that memory fade just because he now understands more than almost anyone.

Ilya Sutskever

Ilya is harder to describe because so much of his influence is structural rather than personal. AlexNet, the scaling intuitions behind GPT, years as chief scientist at OpenAI, all of it sits underneath work that people use every day without knowing his name is attached to it. What stays with me is not any single paper. It is the seriousness with which he treats the question of what happens if this actually works. Founding Safe Superintelligence with no product roadmap and a single stated mission is not a normal move in this industry. It is the kind of decision you make when you actually believe the stakes are as large as you say they are.

Yann LeCun

LeCun taught me that being the loudest believer in the room is not the same as being right, and that being contrarian is not automatically wrong either. Convolutional networks came out of his insistence that structure matters, that you do not need to brute-force everything through scale alone. He has spent the last few years arguing publicly that large language models are a limited path and that the field should be building world models instead, systems that learn physical reality rather than just predicting the next token. He left Meta to build that argument into a company of his own. Whether or not you agree with him, and plenty of serious people do not, it is a useful discipline to sit with someone who refuses to accept the consensus just because it is winning.

Yoshua Bengio

Bengio is the researcher I return to when I want to remember that this field used to be an act of patience before it became an industry. He kept working on deep learning through the years when almost nobody thought it would go anywhere, and that patience is baked into ideas like word embeddings and attention mechanisms that everything since has been built on top of. In recent years he has turned a large part of his attention toward AI safety and governance, and I respect that shift. It takes a certain honesty to help build something powerful and then spend your later years asking hard questions about what you built.

Alec Radford

Radford is less of a public name than the others, and that is exactly what makes him instructive to me. He was behind GPT, GPT-2, CLIP, Whisper, work that reshaped how the field thinks about language and multimodal learning, largely without chasing the spotlight that came with it. There is a lesson in that for anyone building things in a corner of the world far from where the loud conversations happen. You do not need to be the face of the work for the work to matter.

Georgi Gerganov

Gerganov is the one on this list who feels closest to home. llama.cpp did something the big labs were never going to prioritize: it took large language models and made them run on ordinary hardware, on a laptop, on a phone, without needing a data center behind you. For someone building AI in Tanzania, where compute is expensive and infrastructure is uneven, that work is not academic. It is the difference between a good idea staying a good idea and a good idea actually shipping. He reminded me that democratizing access to AI is not a side project. Sometimes it is the whole point.

None of these six agree with each other on how this should go. Sutskever and LeCun are, in some sense, running opposite bets on what the future of AI even looks like. That disagreement is not a flaw in the field. It is the field working the way it is supposed to. I take something different from each of them, and I am still building my own answer to the question they are all circling.

Author: Shadrackovsky