Skip to content
Alex Novara

AI Roadmap

Learning Diary

What was studied each day, for how long, and the notes taken along the way.

View statistics

  1. 1h · 1 session

  2. 30 min · 1 session

  3. 50 min · 1 session

  4. 1h 15m · 1 session

    • Evaluation evening · Meetup

    Weekly review only: the course pace is fine, no changes to the plan.

  5. 1h 30m · 1 session

  6. 3h 30m · 1 session

    • Fictional Retrieval Summit · Conference · Day 3 of 3
  7. 1h 45m · 1 session

    • Fictional Retrieval Summit · Conference · Day 2 of 3

    Rewrote three fictional support prompts with explicit output schemas.

  8. 3h 30m · 1 session

    • Fictional Retrieval Summit · Conference · Day 1 of 3
  9. 50 min · 1 session

  10. 1h 30m · 1 session

    First module of the prompt patterns course: roles, examples and output formats.

  11. 2h 15m · 1 session

    Compared token counts of the same paragraph in three languages.

  12. 45 min · 1 session

  13. 1h 20m · 1 session

    Byte-pair encoding by hand on a toy vocabulary of eight words.

  14. No study today; planned the week and picked the tokenization reading order.

  15. 1h 15m · 1 session

  16. 1h 30m · 1 session

  17. Finished the explainer. Next up: tokenization.

  18. 1h 25m · 2 sessions

    Short session after the late night. Summarised the residual stream in my own words.

  19. 1h 30m · 1 session

    Went through positional encodings and stayed past midnight on the multi-head section.

  20. 1h 55m · 2 sessions

  21. 1h 30m · 1 session

    Started the transformer explainer. Attention finally reads as a weighted lookup:

    • queries ask
    • keys answer
    • values carry the content
    • Year-end agents workshop · Workshop · Day 4 of 4
    • Year-end agents workshop · Workshop · Day 3 of 4
    • Year-end agents workshop · Workshop · Day 2 of 4
    • Year-end agents workshop · Workshop · Day 1 of 4