Stanford Online

Stanford CS193T Thinking with AI | Autumn 2026 | Core Prompting: summary

YouTube summary8 sectionsWatch on YouTube ↗

This is an AI-generated summary of the YouTube video "Stanford CS193T Thinking with AI | Autumn 2026 | Core Prompting" (Stanford Online), made with Samuraize and published by Samuraize. It condenses the YouTube video into 8 titled sections you can read in a couple of minutes, each linking to the moment in the video it covers.

1
Filed under🎓 Education0 comments🍱 Add to trayReport
Study this

Stanford CS193T Thinking with AI | Autumn 2026 | Core Prompting

Stanford Online

切

Core prompting goals 0:05

Core prompting skills begin with knowing what you want from the model. The lecture stays on interactive chat first, not longer agent work, and says that a clear goal helps you get better results, protect your own voice, and shorten the back-and-forth needed to reach a usable draft or image.

切

Direct style and roles 5:00

You do not need to write in a stiff or formal style. The main advice is to be direct, specific, and detailed. The model can also be guided by giving it a role, such as a seasoned real estate agent or a careful software project manager. That changes the voice of the answer, but it does not add new knowledge to the model itself.

切

Set the task clearly 20:00

State the goal, who is speaking, who the audience is, why the task matters, and any limits on length, format, or sources. If you can, define what success looks like up front. If that is too much work, start small and ask the model to interview you first. You can also give your best draft and ask for a few clarifying questions, then revise the scope from there.

切

Trust the evidence 32:00

Do not ask a model to judge its own confidence or explain what happened inside a previous answer. It cannot see those past calculations, and it may still sound certain when it is wrong. When truth matters, ask for evidence instead. Request citations, open them, and check them yourself. For quotes, copy the text and search it, because a model can paraphrase or alter wording.

切

Observability in the chat 40:32

Observability means making more of the work visible. A PDF review failed because Claude read only 20 pages, and the clue was in its own answer. Asking it to report the total number of students would have made the limit obvious. The same idea helps in an interview prompt: have the model echo back a scope of work and let you approve or refine it before it continues.

切

Thinking and context 47:00

Reasoning models do this internally by iterating before they answer. Their chain of thought can be shown as a transcript, outline, or summary, depending on the product and settings. The lesson then shifts to context management: use a new chat for a new topic, because a long mixed thread wastes tokens, narrows focus, and can carry odd fixations or unhealthy patterns forward.

切

Bias and resource use 1:01:32

The speaker says long chats tend to favor the system prompt and the newest prompt, with middle material getting less weight. That is the lost-in-the-middle problem, so the most important details in a long prompt should sit at the top or bottom. They then turn to resource use. Match the model and effort to the job, use text instead of screenshots when you can, and skip an LLM when a calculator or Ctrl F will do.

切

Keep your own voice 1:08:30

The final point is to keep an independent voice. Do not let AI decide your preferences or your goals for you. Also keep your own knowledge from atrophying by asking the model to write down what it did so you can do it yourself next time. The speaker closes by asking what skill you want to keep sharp, and what outside warning system could tell you when AI is quietly dulling it.

AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.

Summarize your own YouTube video

Paste a YouTube link, article, PDF, ebook or slide deck and get a summary like this in seconds. Free to try, no sign-up needed.

⚔️ Try the YouTube summarizer

Discussion

Sign in to join the discussion. Sign in

More from the Bento Box

Browse the Bento Box →

With your OK, Microsoft Clarity and Google Analytics set cookies (data goes to the US) so we can see what breaks. Cookie details