WORTH TESTING · 26 OCTOBER 2026
A real Stanford lecture on LLMs and agents, wrapped in a fake '$100k agent' post
@TechByNagma's post promises a 20-minute agent sold to Anthropic for $100k and '$1.2M a year selling agents'. None of that is in the video. It's actually the opening lecture of Stanford's CS329A course, given by Prof. Aakanksha Chowdhery and Prof. Azalia Mirhoseini (named on the slides), and a slide refers to the Fall 2025 course, so it isn't new. Verdict: a clear, free primer on why few-shot prompting, step-by-step reasoning, reasoning models and agent workflows work; no coding needed to watch, but it's theory, not a how-to. Timestamps are approximate.
Try this
- Add two or three worked examples to a prompt and compare the result with no examples. 05:00
- Ask the model to reason step by step before answering logic or maths questions. 06:40
- Don't expect step-by-step prompting to help much on small models. 08:35
- Use chat (instruction-tuned) models rather than base models when you want instructions followed. 14:30
- For hard tasks, generate several answers and let a checker, such as tests, pick the best. 19:15
- When generating several answers, raise the temperature to get more variety. 21:28
- Use a reasoning model for complex multi-step problems and a standard one for simple tasks. 24:14
- Give an agent a clear end goal and let it plan, act, check and correct. 40:18
- Start with prompt chaining: draft, then critique, then revise, as separate steps. 41:00
- Where nothing can be checked automatically, have a second model judge the output against a rubric. 46:30
Timestamps point to the moment in the source video.
Cost
Free lecture.
Source
Hype or Fact?
Hype A '$100k agent sold to Anthropic' post wrapped around a Stanford lecture
How we knowSteps to try, not a claim that we independently tested every result. Follow the linked sources and check current prices. Our methods & sources ↗