Uploaded June 2026 | Updated September 2026, 7 hours ago
Training an AI model can sound like something reserved for large research labs, expensive GPU clusters, and datasets scraped from half the internet. But what does “training” actually mean? How much data do you really need, and when does a smaller, better dataset beat a massive one?
In this live stream, we’ll talk with Piotr Miłkowski about how developers can start teaching models in practice. We’ll look at what makes a dataset useful, whether AI memorizes or understands, how models learn from mistakes, and why data from places like TikTok comments, Reddit, or Wikipedia can produce very different outcomes.
We’ll also cover the practical side: common beginner mistakes, what a single developer can realistically train today, when it makes sense to train your own model instead of using an API, and what you can build over a weekend with a GPU.
Training an AI model can sound like something reserved for large research labs, expensive GPU clusters, and datasets scraped from half the internet. But what does “training” actually mean? How much data do you really need, and when does a smaller, better dataset beat a massive one?
In this live stream, we’ll talk with Piotr Miłkowski about how developers can start teaching models in practice. We’ll look at what makes a dataset useful, whether AI memorizes or understands, how models learn from mistakes, and why data from places like TikTok comments, Reddit, or Wikipedia can produce very different outcomes.
We’ll also cover the practical side: common beginner mistakes, what a single developer can realistically train today, when it makes sense to train your own model instead of using an API, and what you can build over a weekend with a GPU.
