> ## Content Index
> Fetch the complete content index at: https://longtalkshort.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Never Open With a Joke, Never End With "Thank You": MIT's Legendary Rules for Speaking
- URL: https://longtalkshort.com/patrick-winston-how-to-speak-mit-lecture-summary/
- Published: 2026-10-26T00:16:31.000Z
- Updated: 2026-10-26T00:16:31.000Z
- Description: Your first sentence is probably a joke. Your last is probably "thank you." MIT's Patrick Winston says both are mistakes, and his fixes have drawn more than 23 million views.
- Author: Truman
- Tags: Psychology, Patrick Winston

**You probably open your talks with a joke and close them with "thank you."**

According to one of MIT's most-watched lecturers, **both are mistakes.**

For more than 40 years, computer scientist Patrick Winston gave a talk called "How to Speak" every January. He ran MIT's Artificial Intelligence Laboratory from 1972 to 1997 and died in July 2019\. The version MIT OpenCourseWare filmed in 2018 has since passed 23 million views.

It isn't about charisma. **It's a set of rules anyone can learn.**

*In short: speaking well depends mostly on knowledge and practice, not talent. Open with a promise, repeat and mark your structure, keep slides sparse, show passion, and finish on your contributions and a salute instead of "thank you."*

## 1\. Talent barely matters

Winston opens with a formula: quality of communication depends on knowledge, practice and talent, with the talent term "very small."

His proof is a story from Sun Valley. On a celebrity weekend, he spotted Olympic gymnast Mary Lou Retton on a beginner slope and realized **he skied better than she did.** She had the talent. He had the knowledge and the practice.

He also bans laptops and phones. We have "only one language processor," he says, and email distracts you and everyone near you.

So if skill is learnable, where does it start? With your very first sentence.

## 2\. Start with a promise, not a joke

At the start, people are still settling in and adjusting to your voice. **The joke falls flat.**

Open instead with an **"empowerment promise"**: what the audience will know at the end that they didn't know at the start.

Then keep four habits running through the whole talk:

- **Cycle.** Say the key idea three times. At any moment, about 20 percent of the room is "fogged out."
- **Build a fence.** Show what your idea is not, so no one confuses it with someone else's: "mine is linear, his is exponential."
- **Use verbal punctuation.** Number and announce your sections so people who drifted can "get back on the bus."
- **Ask a question.** Not too easy, not too hard. Then wait about seven seconds for an answer.

![Four cards: start with an empowerment promise, not a joke; in the middle cycle, build a fence, add verbal punctuation, ask a question and wait; end on contributions; last words a joke or salute.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig1-16.png)

The four moments Winston plans for, from his first sentence to his last. Concept diagram. Made for Long Talk Short.

Structure is half of it. The other half is the room you speak in, and what you put on the wall.

## 3\. Pick your room, and kill most of your slides

His best time to lecture is 11 a.m., when people are awake and not yet sleepy after lunch. Keep the room well lit, because dim light signals sleep:

> it's extremely hard to see slides through closed eyelids

Check out the room beforehand, and pick one that will be at least half full.

**To teach, he prefers the board to slides.** Writing goes at roughly the speed people take in ideas, lets you draw, and gives your hands something to do besides hiding in your pockets.

Props stick in the memory too: a spinning bicycle wheel that explains gyroscopes, or a physics professor who let a heavy pendulum swing back and stop just short of his nose. Students, he notes, always ask for "more chalk, less PowerPoint." His "lunatic fringe" explanation: watching someone write or handle an object triggers "empathetic mirroring."

**Slides are for exposing ideas, not teaching them.** His usual verdict: "too many slides, too many words." His fixes:

- Cut the logos, the background clutter, the title and most of the text.
- Use big fonts, around 40 points.
- Skip the laser pointer. Aiming it turns your back to the room, so draw an arrow on the slide instead.
- Print the deck and lay it out on a table to see if it's too heavy.

Getting ideas across is one thing. Getting people to act on them is another.

## 4\. Inspire, then persuade

Winston surveyed people at MIT about who had inspired them. Incoming freshmen remembered a teacher who told them they could do it. Senior faculty remembered someone who helped them see a problem in a new way. **Everyone remembered someone who showed passion for their work.**

To persuade, situate your work: the problem, its context and why it matters. Practice with friends who don't know your work, because colleagues and advisors "hallucinate" material that isn't there.

In a job talk, two colleagues on a Navy science board told him, a faculty candidate has five minutes to show two things: **a vision, and that they've done something.**

- **A vision** is a problem someone cares about plus a new approach.
- **Proof you've done something** is the list of steps toward the solution, ending on the contributions that match them.

Even a convincing talk fades, though. Winston had a five-point recipe for making it stick.

## 5\. Get remembered, then stop well

Memorable work needs **"Winston's star"**: a **symbol**, a **slogan**, a **surprise**, a **salient** idea that stands out, and a **story**.

His 1970 PhD work had all five by accident: the arch, "one-shot learning," the surprise that one example can be enough, the "near miss," and the story of how it worked.

![A gold five-point star surrounded by five cards: Symbol, the arch; Slogan, one-shot learning; Surprise, you can learn from one example; Salient idea, the near miss; Story, how it works and why it matters.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/10/fig2-16.png)

The five S's Winston said his arch-learning thesis had by accident. Concept diagram. Made for Long Talk Short.

Then the ending, where most talks fall apart:

- **No collaborators slide at the end.** It implies you did a thousandth of the work. Credit people on the first slide.
- **No "Questions?", "The End" or web address.** They waste the slide that stays up longest.
- **End on "contributions."** Conclusions are fine, but contributions are what people care about.
- **Now a joke can work,** because the audience is used to your voice.
- **Salute, don't thank.** "Thank you for listening" is "a weak move" that suggests people stayed out of politeness. Say what you valued about being there instead. His example is Bill Clinton's 2012 convention speech, which ended on "God bless you and God bless America."

*Struggling more with nerves than structure? Our piece on* [*Vinh Giang's way out of shyness*](https://longtalkshort.com/vinh-giang-shy-is-a-habit-record-and-review/) *covers the delivery side.*

## Weighing it: what holds up

**The laptop ban has research behind it.** A 2013 study by Sana, Weston and Cepeda found students who multitasked on laptops scored 11 percent lower on a comprehension test after a lecture, and students who simply sat in view of them scored lower too.

**Wait time is real, but seven seconds is his own number.** Mary Budd Rowe's classroom research found teachers typically waited about one second for an answer, and that waiting three to five seconds produced longer, more thoughtful replies.

**Some rules are rules of thumb.** His slides experiment, a student teaching freshmen in his fraternity, is an anecdote with no published study behind it. The "20 percent fogged out" figure and the 11 a.m. advice are his own heuristics.

**Mirroring is a guess, as he admitted.** Mirror neurons are well established in monkeys, but whether they explain how humans understand actions is still disputed.

**Two small details are off.** The Dartmouth AI proposal he shows was written in 1955 for a 1956 summer project, not 1957\. Clinton's 2012 speech was a nominating address; Julián Castro gave the keynote.

## Long Talk Short

- Speaking well comes mostly from knowledge and practice, not talent.
- Open with an empowerment promise: what people will know by the end.
- Cycle the key idea three times, fence it off, number your sections and ask real questions.
- Teach with the board and props; use sparse slides with big fonts and no laser pointer.
- In a job talk, show a vision and that you've done something, fast.
- Make work memorable with a symbol, slogan, surprise, salient idea and story.
- End on contributions and a salute, not "thank you for listening."

**The one thing to remember:** your first and last moments matter most. Promise something at the start, and leave up what you did at the end.

Source: ["How to Speak"](https://www.youtube.com/watch?v=Unzc731iCUY&ref=longtalkshort.com), Patrick Winston, MIT OpenCourseWare (RES.TLL-005, recorded during MIT's Independent Activities Period, January 2018). This is an independent summary and is not affiliated with MIT, MIT OpenCourseWare or the Winston family.

*This article is for general information only. Quotes are from the lecture's automatic captions and may contain small transcription errors.*