Back to index
OtherSelf-paced, 40+ hours of recorded workshops and talks, organised into evals, RAG, building applications, prompt engineering and fine-tuning·Free at parlance-labs.com/education; the original Maven listing charged a nominal $20 and is now closed

Mastering LLMs: Open Course (Hamel Husain, Dan Becker and Parlance Labs)

4.4

One of the best free collections of practitioner LLM teaching ever assembled, recorded during a short window in 2024 when almost everyone worth hearing showed up. Some of the fine-tuning material has aged, but the evals and RAG talks still hold up. Treat it as a library, not a course.

What We Liked

  • Completely free, with notes and slides alongside each talk
  • Remarkable speaker list: Jeremy Howard, Simon Willison, JJ Allaire, Wing Lian, Eugene Yan, Shreya Shankar, Jason Liu, Charles Frye and more
  • Talks are honest and opinionated, including one literally titled Why Fine Tuning is Dead inside a fine-tuning course
  • Organised by topic with a sensible order for fine-tuning: should you, how to, deploying, then advanced
  • Early versions of ideas that became Hamel and Shreya's evals course and Jason Liu's RAG course, for free

What Could Be Better

  • Recorded in mid-2024, so model names, tool versions and some fine-tuning advice are now dated
  • Fine-tuning dominates the catalogue at the expense of agents, which barely appear
  • No support, assessment or certificate; the Maven version is closed and never offered office hours for late joiners
  • Quality and audio vary between talks, as you would expect from a conference turned course

Detailed review

In May 2024 Hamel Husain and Dan Becker set out to teach a short course on fine-tuning LLMs. Within days it had grown into a multi-week conference with more than 2,000 participants and a speaker list most paid events could not assemble. Two months later they published the whole thing for free, organised and annotated, at parlance-labs.com/education. That page is what this review covers.

The content is grouped into evals, RAG, building applications, prompt engineering and fine-tuning, with fine-tuning split into four parts: whether you should do it at all, how to do it, deploying fine-tuned models, and advanced topics like napkin math for memory, FSDP, DeepSpeed and function calling. Husain's write-up puts the total at more than 40 hours, taught by over 25 practitioners including Jeremy Howard, Simon Willison, JJ Allaire, Wing Lian of Axolotl, Mark Saroufim, Zach Mueller, Eugene Yan, Shreya Shankar, Jason Liu, Charles Frye, Kyle Corbitt and Ankur Goyal. What I like most is the honesty. The fine-tuning section opens with talks on when you should and should not fine-tune, and one of them is titled Why Fine Tuning is Dead.

That kind of internal disagreement is far more useful than a single instructor's confident line. The evals and RAG sections are the strongest today. You can see early forms of the frameworks Husain and Shankar later turned into their evals course, and Liu's talk on systematically improving RAG is effectively a free preview of his paid cohort. If you are deciding whether to spend money on either, watch these first.

The weaknesses come from the date. Everything was recorded in 2024, before agentic coding tools and reasoning models changed how most teams work. Specific model recommendations, library versions and parts of the fine-tuning advice are stale, and agents barely feature. There is also no support.

The old Maven listing, priced at $20 only because Maven does not allow free courses, is now closed and said outright there were no office hours or questions for later students. So use it as a reference library. Pick the talk that matches your current problem, watch it at 1.5x speed with the notes open, and move on. Free practitioner material of this quality is rare, and it is still worth an afternoon of anyone's time.

[ final ]

The verdict.

Bookmark it and pick talks by topic. Start with the evals and RAG sections, then the Should you fine-tune? block, and only go deeper into fine-tuning if you have a clear reason to. For the price, nothing else comes close.