This video is the first lesson in a course on the "Solveit" method, a philosophy and platform for problem-solving with code and AI, presented by Jeremy Howard and Jonno of FastAI. The core idea is to use AI not to outsource work, but as a collaborative partner to enhance one's craft and deepen understanding. The lesson begins with community logistics, emphasizing the role of their Discord server. Howard then demonstrates the Solveit method through three distinct examples: first, by iteratively solving an Advent of Code puzzle, showcasing an exploratory, step-by-step coding process; second, by writing a high-quality article from an interview transcript, treating it as a "style transfer" problem to avoid generic AI-generated text; and third, Jonno builds a live, interactive web application from scratch within the Solveit environment, highlighting how changes in the dialogue are instantly reflected in the running app. The overarching theme is a focus on small, verifiable steps, continuous learning, and creating well-crafted, human-driven solutions.
Lesson 1: The Solveit Method for Coding, Writing, and Building
0:40 Community and Logistics: The lesson begins by outlining the course's community structure, primarily on Discord. Key channels for announcements, introductions, sharing work (share-it), and getting help are introduced. The !ask and !ask_sassy Discord bots are presented as tools for getting AI-powered assistance.
15:56 The Solveit Philosophy: The core principle is not to outsource tasks to AI, but to use it as a tool to improve your skills and deepen your understanding. This "human-agent" collaboration focuses on learning and improving your craft, whether it's coding, writing, or problem-solving.
23:41 Solving Problems Worth Solving: Howard argues against using AI to simply generate boilerplate code. Instead, the goal should be to eliminate boilerplate by creating elegant, concise, and reusable abstractions. He demonstrates this with a real-world example from the Solveit codebase, where he replaced a complex library with a simple, custom function.
31:23 Example 1 - Coding with Advent of Code: The first practical demonstration involves solving an Advent of Code puzzle.
(39:02) The process is highly iterative: load a small piece of data, inspect it, apply a single transformation (like splitting lines), and inspect the result again before moving to the next step.
(47:27) The AI is used as a learning partner to ask for more concise or idiomatic ways to write Python code, rather than just asking for the final solution.
(52:42) The final, working code is assembled from the small, verified steps taken during the exploration, which also serves as documentation for the process.
(59:54) Even a simple problem is an opportunity to learn. The problem is solved a second time using NumPy to demonstrate how to practice and apply new skills.
1:05:25 Example 2 - Writing a High-Quality Article: Howard showcases how he wrote a well-received article based on a transcript of his interview with Chris Lattner.
(1:08:21) The key is to avoid starting from a blank slate. By using a transcript, the task becomes "style transfer"—refining existing, authentic content into a new format, which avoids generic "AI slop."
(1:12:08) The process involves having the AI generate drafts for small sections, which are then heavily edited or completely rewritten by the human. Editing the AI's output directly teaches it your preferred style, leading to better suggestions over time.
1:21:22 Example 3 - Building a Live Web App: Jonno demonstrates building a dynamic web application where users can vote on AI-generated images.
(1:30:52) The application is built incrementally within a single Solveit dialogue, starting with API calls, then generating images, and finally creating the user interface with FastHTML.
(1:36:33) The AI is used to assist with styling (CSS), with screenshots of the current UI provided as context for better results.
(1:51:36) A key feature is that the web app runs live directly from the dialogue. Any code changes are instantly reflected in the public-facing application without needing a separate build or deployment step.
1:54:04 Key Takeaway and Homework: The lesson concludes by emphasizing the power of working in small steps with rapid, direct feedback. Viewers are encouraged to practice this method by tackling a small project, such as an Advent of Code puzzle or turning a favorite video into an article, and sharing their work with the community.