Class 3: August 25, 2026 · ENTR 5000
Notes from August 25, 2026. These notes record the discussion and instructions given that day. For current requirements and due dates, use the assignment guidelines and semester schedule.
Building expands what becomes possible
The adjacent possible describes how a new skill or working experiment exposes nearby opportunities that were previously hard to see. In class, it was compared to a branching skill tree: building one small tool can reveal the next useful tool or a new combination. AI can make this cycle cheaper, but the learning comes from examining what happened and trying the next step. Steven Johnson’s explanation provides context.
AI use can support continuous learning when you ask questions, examine how a solution works and apply the knowledge again. The discussion connected this to James Clear’s identity-based habits: seeing yourself as a curious builder can support repeated experimentation.
A practical understanding of models
We revisited 3Blue1Brown’s neural-network introduction. In the digit-recognition example, pixel values enter the network, weighted connections and biases transform values across layers, and output values indicate the predicted digit. Training adjusts parameters using examples; using a trained network applies those learned settings to an input. The lesson’s purpose was conceptual literacy, not deriving the mathematics.
A deterministic procedure produces the same result from the same input and conditions. An LLM can instead generate different plausible responses. This distinction matters when a task needs exact constraints: interpreting a scheduling request in natural language is useful, but actual scheduling calculations should be executed and checked with appropriate tools rather than trusted because the generated text looks reasonable.
From a demonstration to a usable application
A local prototype runs on your computer. Remote hosting makes an application reachable from other devices and can keep it operating when your laptop is closed. A working screen, a live data connection, reliable hosting and the ability to support customers are different milestones.
Containers were introduced as a way to package an application’s environment and reduce “works on my machine” differences. Docker’s container explanation is a useful reference. A demonstration that works once is not automatically a product ready to sell and maintain.
Before Thursday, August 27
- Post the link or screenshot from your initial build if you have not yet shared it; complete your Slack profile.
- Bring ideas for what to explore in a three-week sprint: a product to build, a next step for an existing venture, or a question you want to answer. Working alone or with classmates is possible.
- Be ready to discuss goals and continue show-and-tell. No additional required video was assigned in the recorded closing.