Companion videos

Chain-of-thought, self-critique, tree-of-thoughts — companion videos

The article sorts these techniques by the kind of problem they actually help with, not by which sounds most impressive. These videos give you the same picture from two angles: an Anthropic walkthrough of building a structured, thought-out prompt from scratch, and the original Tree-of-Thoughts paper explained by someone who understands the search behind it.

Primary pick

24:52

Prompting 101

Anthropic

Two members of Anthropic's Applied AI team build up a real-world prompt live, layering in role, structured sections and explicit thinking steps. It's the most credible "how to use chain-of-thought without ritualizing it" demo on YouTube, and it makes the article's "stop pasting Let's think step by step into everything" point concrete.

What you should get from this: You can layer role, structured sections and explicit thinking steps into a prompt without turning chain-of-thought into a ritual.

Watch or know first: Solid basic prompting; the article's technique taxonomy is the right frame going in.

AI Expert note: Chosen because it shows thinking steps used inside a real prompt build — the antidote to pasting 'let's think step by step' everywhere, which is the article's core warning.

Open video page

Also worth watching

29:29

Tree of Thoughts: Deliberate Problem Solving with Large Language Models (Full Paper Review)

Yannic Kilcher

Yannic walks through the original Tree-of-Thoughts paper from Princeton/DeepMind — the generate-score-prune-backtrack loop, why it lifts Game of 24 accuracy from 4% to 74%, and where the technique stops being worth its 10× token cost. Useful precisely because it forces you to see ToT as a search algorithm, not a magic prompt prefix.

What you should get from this: Understand generate-score-prune reasoning patterns and when they are too expensive for real work.

Watch or know first: Comfortable with technical paper walkthroughs; you don't need to have read the paper.

AI Expert note: Included to make ToT concrete as a search algorithm with a real token bill — the 10x-cost boundary is the practical takeaway, not the benchmark numbers.

Open video page