Curated lists and roadmaps
Other people have already organized this field. Use their lists as a map and check what is current.
Words on this page (3)
- repo
- : Repository: a project folder tracked by git, usually hosted on GitHub.
- agent
- : A model run in a loop that chooses its own tools and steps until the task is done.
- tool
- : A function the model can ask your program to run, described by a name, description and argument schema.
How to use a list
Other people have already sorted this field into lists. A good list saves you weeks. A bad one is a pile of bookmarks. Judge a list in a minute with four checks:
When was the last update? A list untouched for a year is probably stale. Does each entry say why it is worth reading, or is it just a link? Is it maintained by one person or many? Are paid items marked?
A worked example: judging a list
Open the and look at the commit history. If the last change was eleven months ago, treat the links as hints and check each one. Look at ten entries. If seven of them have a sentence about what you will learn, the list is curated. If none do, it is a dump. Pick the best two or three entries in the section you need and ignore the rest.
Which list for which goal
You are new and want an order to follow: the path on this site, then the beginner guides below. You want to build : the agentic engineering lists. You want to study the research: the model internals chapter and the papers it cites. You want to keep up weekly: the sources in the previous lesson.
Use this site's library
The Resource library page lists every link cited in these lessons, with a sentence on what each one is good for, filterable by type. It is smaller than a giant list and every entry is annotated.
What goes wrong: link rot and repeats
Open ten random entries on a list. If three lead to an error page, a changed address, or a site that no longer covers the topic, the list is not being maintained. This is link rot, and it grows with age.
A second problem is repetition. Open three lists on the same topic and you often see the same ten resources in each. The real count is not 500 links across three lists. It may be closer to 40 unique ones. To confirm, copy the titles into one document and remove duplicates. The number left is the size of the topic.
Turning a list into a plan
A list becomes useful when you add a goal and a limit.
- Write one goal, such as "build a small agent that calls two ".
- Find the section of the list that matches and copy five entries into a notes file.
- Next to each, write one line on why you picked it.
- Spend 15 minutes on each. Cross out any that repeat what you already learned.
- Keep the best two and do them this week.
Five entries become two in about an hour and a quarter. Expect to end with fewer items than you started with. That is the point.
Common mistake
Saving lists you never open. Pick one, take three items, and read them this week.
A beginner opens a list titled "500 AI resources" and closes it ten minutes later more confused. Long lists are libraries, not courses. Pairing a list with a path that says "read these five first" is what turns links into learning.
Check yourself
What makes a list entry more useful?
Not quite. Context tells you whether to read it.
What does a list of 500 links give you?
Not quite. A curated order is what turns links into a path.
A list was last updated 14 months ago. What should you do?
Not quite. Lists age quickly in this field, so check recency.
You got 0 of 3 right.
Go deeper
- Repo
EthicalML · Articles, courses, papers and benchmarks, with monthly updates.
- Repo
keyuchen21 · A phased roadmap of 200+ resources.
- Repo
fatihkc · Focused on directing coding agents.
- Repo
Machine Learning Tokyo · Ten free resources ranked by quality.
- Repo
Louis-François Bouchard · A guide for beginners organized by learning style.
- Repo
Maxime Labonne · Roadmaps for LLM scientists and engineers with notebooks.
- Repo
Microsoft · Structured lessons with code.
- Tool
roadmap.sh · Community map of skills.
Spotted a mistake or something unclear? Report it on GitHub.