1.1.5. Google Colab — Your AI Experimentation Lab

Google Colab overview (how to run the program’s Python code)

• Google Colab as a “lab”: It’s positioned as the place where you can execute Python code from the course/materials.
• Access URL: collab.google.com
• Create a notebook: Click “New notebook” to get an empty code cell.
• Workflow: Copy/paste code → click the Play/run button → execute.
• Add/manage cells:
  • + Code to insert additional code cells
  • Trash icon to delete a cell
• Save files: Codes can be saved to Google Drive or GitHub, and notebooks can be named (e.g., “my first Python code”).

Demonstrated code execution concepts

• Basic print() behavior: Examples show simple output like printing “Hello …” and repeating output (e.g., printing “50 times” when using # comments correctly or incorrectly).
• Comments vs execution (Python #):
  • If a line was meant to be a comment but is missing the leading #, Colab treats it as code → can cause SyntaxError.
  • Lines become green when they’re treated as comments; the speaker notes that commented-out lines won’t execute.
• Common student mistakes:
  • Missing a character such as # at the start of a comment.
  • Copy/pasting extra lines that lead to errors.
  • Syntax issues like missing parentheses.
  • Error highlights appear via red underlines for problematic code.

Using Gemini inside Colab to handle errors and modify code

• Gemini integration:
  • Colab includes Gemini for explaining error messages (e.g., “Invalid syntax” / “Syntax error”).
  • The speaker warns results are not guaranteed: sometimes Gemini’s explanations are confusing or suggestions don’t fix the code.
• Fixing inside Colab:
  • There’s an “explain” and a “fix this code in the last cell” flow.
  • When it works, it can automatically correct the faulty line so the notebook runs successfully.
• Using Gemini output elsewhere:
  • If Colab’s Gemini is weak, the speaker recommends copying the code + error message and pasting into another AI tool (mentions ChatGPT / cloud AI).
  • The speaker claims Cloud AI (and possibly newer Gemini like “Gemini 2.5 flash” mentioned) may be more capable for code fixing than the built-in one, though they admit it’s subjective.

Larger example: randomization and functions

• The video includes an example that creates a face using special characters.
• It uses the Python random library:
  • Each run produces a different face because it randomly selects from multiple eye and mouth options (e.g., 6 possible eye shapes and 6 possible mouth shapes).
• Function definition and calling:
  • A function is defined (e.g., make face) and then called.
  • Calling the function multiple times (e.g., running it twice) prints multiple faces.

System requirements / environment details

• Need a Gmail account:
  • Gmail is presented as required to access Google Colab, Gemini, and Google Drive.
• Runtime/GPU settings:
  • Default runtime is CPU.
  • For AI model training, you may need to switch to T4 GPU via Runtime → change run type.
  • The speaker notes the process may require:
    • Disconnecting and deleting/reconnecting the runtime to obtain GPU access.
• Resource monitoring:
  • Mentions viewing RAM and disk resources, including temporary space and memory.
• GPU availability limits:
  • Sometimes Colab denies GPU access due to usage limits.
  • The speaker notes messages like being unable to currently connect to GPU, and suggests waiting and retrying later.

Learning approach emphasized

• Philosophy: “Copy, paste, run, and learn.”
• When stuck: use Gemini inside Colab, or copy/paste to an external AI partner and ask with the error message + code.
• Encourages experimentation: tweak the code and explore to learn.

Main speakers / sources

• Single course presenter (no named guest speakers in the subtitles)
• Google Colab / built-in Gemini
• External AI partners mentioned: ChatGPT and “cloud AI” / “Google cloud AI”