About me - Krzysztof Sopyła, PhD

I am Krzysztof Sopyła, an AI R&D lead and researcher with a deep curiosity about how reasoning works, how intelligence emerges in computer systems. My work primarily revolves around efficient and effective ML systems that can learn and reason with drastically less data and compute. I don’t believe that the current way of training LLM and LRM is the optimal one. It worries me that we are wasting the earth’s resources, blindly building AGI using brute force, there should be a better way.

Over the years, I’ve had the opportunity to work on various projects involving natural language processing and neural networks training. During my PhD I was working on online learning methods and GPU-accelerated algorithms (especially SVM). After my university years I have decided use my skills more practically and I have worked as AI lead in a few startups and companies: woking on Goodwrite a polish writing assistant, were in 2018 we started using the brand new Transformer architecture, then I’ve contributed to the development of one of the first Polish BERT models. I and my team try to understand the book content and we were working on extending the context of BERT models, as well as LSTM for for book recommendation engines. Currently, I’m exploring conversational tutoring systems at Pearson.

I like revisiting old methods, ideas and concepts, mixing them with modern enginnering practice and architectures. I believe in experimentation and reproducibility of reserach. Even small well design experiment could lead us in proper direction.

One of my core beliefs is education and knowledge sharing. Teaching and learning was always important to me, it gives me purpuse and make me feel fulfilled. Sharing code, data and ideas, having the oportuninty to discuss them in respectfull manner lightens me up. I’m working on a PyTorch course where my Prof. Torchenstein (mad scientist) will teach you how to use PyTorch. The goal is to make Pytorch learning fun as reading a comic book.

I’m always eager to learn, collaborate, and revisit old concepts with fresh perspectives. If you share an interest in AI research or efficient system design, feel free to reach out!

You can find me on GitHub, LinkedIn, Google Scholar, Twitter. I used to write about AI in Polish on my blog at ksopyla.com, where I’ve documented many of my earlier explorations.