Today: Last lecture, your future in CS, conclusions

Thanks To Ngoc and Iddah and Section Leaders

Thanks to Ngoc and Iddah and the section leaders! The only way this course can work is with their prodigious and generous efforts. The staff, working in the "CS198" program, are selected for technical skill and teaching generosity - a fantastic group of people and we are lucky to have them.

Final Exam

Python Guide

I'm gradually working on and expanding the Python Guide, aiming to keep it as a free resource on the web. If you want to find it in the future, it's linked from my home page and the CS106A page.

What is the Role of CS106A?

CS106A First Day Claims vs. Today

You will never not know this nature of the computer. Even if you never write another line of code.

The natural rest state of code is "broken". And yet you work on it iteratively a few times, and then suddenly it works, and there's that feeling of accomplishment.

alt: dogs try again and again and it does not work, 'it will work, you will see' they say

Learned All The Programming Techniques?

Sadly, no.

Learned the Important Core

Here is the deal: Python and the space of all programming techniques is very large. A bigger space than you might think. The "Programming Python" book is over 1600 pages long. Fortunately, many of these features are for rare cases, you do not need to know them to get things done.

You have learned the most important 80% core: loops, lists, strings, functions, tests, files. There's a few more important techniques in CS106B, in particular, recursion.

In some AI-aided world, you will need to know these code fundamentals to work with the AI, so CS106A has set you up with those basics.

Building on these core concepts, you know enough to read about (or work with the AI), to solve more difficult problems.

AI and the Future of Programming

I talk about programmer shortage first, and then AI coming on the scene.

Historic Programmer Shortage 1990-2026

Why is there a Programmer Shortage?

Nick Python T-Shirt Story

I was on a bicycle, wearing ratty clothes and a "Python" t-shirt stopped for a red light. A person walking in the cross-walk in front of me, stopped, turned to me, and asked if I was looking for work.

Not to disillusion you about graduating from Stanford, but that is not how hiring is normally done.

Like how desperate for programmers was that person? That is what an extreme programmer shortage looks like!

Free Breakfast Story

When I worked at Google, there was a nice free breakfast available at the office. The free breakfast would end at 10am. The funny part is that some people would complain .. the free breakfast cuts off too early, how am I supposed to get here by 10am?

What does this story tell you? That google cares about the morning nutrition of its team? Or that google is spending money on entitled nerds in the face of a desperate programmer shortage.

I suppose the other lesson is that it is human nature to grow accustomed to whatever blessings one has in life, and then risk seeing them as entitlements.

Let's talk about AI and programming, and the range of how it might work in the future.

Preface: AI Land Grab

Why is there so much hype and spending by companies to get into AI? I think there is a rush-to-occupy-new-land mentality, as by getting there first, they can dominate the space for the future. This suggests spending huge sums to be first. It is a valuable new space, but I'm skeptical that these vast sums will pay off.

Preface: Much Code is Formulaic

Much computer code is formulaic, resembling code that has been written before. Also running code often has tangible outputs, so this all makes code a suitable area for AI. Contrast to, say, writing a movie script. How does the AI know if the script is good? The domain is muddy and difficult.

Hybrid - Programmer Working With AI

Think about a human programmer working with an AI to solve problems more quickly.

1. AI Autocomplete - 2x

The AI could work as a fancy autocomplete, proposing and finishing work as the programmer works on each section of code. This works today, so this is the lowest estimate of speedup.

2. AI / Programmer Interactive Dialogue - 10x

Suppose the programmer suggests the problem space, then the AI suggests an outline of the code strategy. The programmer provides feedback suggestions. The AI write some code. The programmer surveys the code and provides nudges and corrections. The two go back and forth. If the AI is writing nearly all the code, this could be dramatically faster than the programmer writing all the code. The programmer is providing advice and corrections, but not looking at all the code lines in detail. We'll call this "10x" faster, more to give it a name than make an exact numeric prediction. AI systems today are approaching this level, but still making mistakes that require human oversight, looking at the code.

3. AI Maximalist - e.g. AI iPhone app

The maximum AI story is one where the programmer really does not need to look at the AI produced code. The defect rate in the code is so low, we just trust it. (This is not how things work today, but maybe someday.)

Suppose in his model a programmer wants an iPhone app, describing the app to the AI. The AI produces a prototype app, and the programmer runs it to see how it works. The programmer provides feedback about desired changes, and the AI does another version, and so on.

The role here of the "programmer" is as a source of "taste" and user needs, not the code. This is akin to a Hollywood director giving feedback to the actors and camera operators. The director does not run the camera, but is nudging the process with their high level vision of what they want for the final product.

Story: Less Demand for Programmers

It's possible that AI code writing will reduce demand for programmers. Programmer wages have been elevated for years, so the wages could come down. Society could still function fine without CS being more highly paid than other professions. Data point: the Stanford 2025 CS graduates got jobs, although anecdotally they had to work harder to get them.

Counter: AI Screwups

It's well known that AI models can make real mistakes. There are stories of the AI code including some big screwup - here is a recent Amazon example, but there is nothing Amazon specific about this pattern.

If a programmer needs to look at all the lines of code, the speedup will be limited. On the other hand, I expect the AI will gradually get better at avoiding screwups.

Counter: Jevon's Paradox

Jevon's Paradox. The AI is making the production of software cheaper. A well known effect of this is to increase the amount of software. So in effect, they increased productivity may be absorbed by making a greater volume of software.

Counter: Accountants vs. Spreadsheets

At one time, accounts did a lot of arithmetic, literally adding and balancing sums. With spreadsheets and software, computers now do the arithmetic .. and yet accountants still exist. Now accountants work on higher-level concepts of strategy and legal compliance. Like a computer system, their work is "on top of" the basic math done by the computer.

However, sometimes a job is truly automated away. There used to be lamplighters to light and manage flame lamps outside. But electric lighting has certainly caused that job to vanish. (See also: whale hunting).

Aside: the history of artificial light from prehistory to today is a neat example of the development and economics of technology over time. See: History of Illumination (20 minute podcast), and the book "Brilliant" by Jane Brox (amazon link)

Computers are Valuable

Nobody knows how this will work out. My gut is that computers are a very valuable technology, and so jobs working with computers in some way or another will continue to be numerous and important. That said, this is clearly a time of high variability, so we'll have to see how things work out.

Editorial - Choosing a Major

Think about choosing a major - Nick's hand-wavy editorial

Background: Many Computer Languages

Python Niche - Programmer Efficient

Code Ideas We've Seen in Python

Your Second Programming Language

Here is some C++ code 

// comments start with 2 slashes
int i = 0;                 // declare var + type
while (i < 100) {          // parens + braces
    i += 1;                // same as py + semicolon
    if (is_bad(i)) {       // parens + braces
        return;
    }
    i += "Hello";          // error detected
    // int/string types different,
    // so above does not work.
    // Error is flagged at edit-time:
    // earlier than python, an improvement
}

Possible Next Steps

Most Stanford students take 1 or 2 CS classes and keep with their chosen major. It's easy to imagine they use Python here and there as part of their work.

After CS106A ..

Next "CS106" CS106B

Aside: Selecting A Major

Selecting a major, it's nice if you enjoy the sort of work that is a big part of that major. CS106B plays this role a bit for CS.

How do you feel, finishing each CS106B project. There are many ways to pick a major. One angle is - you should enjoy the main topics that make up a major.

Think About Section Leading

CS Major - Code and Math Tracks

CS has programming, as you would expect. There is also a parallel CS-math track, feature integer mathematics, proofs, and reasoning with probability distributions. 

  CS Major - Parallel Tracks
Programming          Math
CS106A               CS103
CS106B               CS109
CS107                CS161
CS111

Next Course FAQ

Aside: What is CS Integer Mathematics?

CS Major Tracks / Concentrations

Some Select Courses

We'll just mention a few courses you could take, build the picture that there are many different areas of CS you might explore. Many of these require CS106B as the pre-requisite.

Scientific Python CME 193

Applied Machine Learning CS129

Human Computer Interaction (HCI) CS147

Graphics CS148

Human Computer Interaction - HCI Design

Handle examples - the point here, is that the object communicates its function visually, and this connects to some low-level part of the brain, so you just know intuitively how to operate it. The best interfaces work this way, and the language of it is visual, not words.

This last one is wrong. It's a push door, but they have put a "pull" handle on it, so people keep using it the wrong way. Putting the "push" sign on their is not the right solution.

Symbolic Systems Major

A sibling to the CS major - similar intellectual domains but less focus on coding

An interdisciplinary major that uses the lenses of CS, Philosophy, Psychology and Linguistics to study systems that use symbols to represent information. In Symsys you can concentrate on AI, Neuroscience, Natural Language, Philosophical Foundations or design your own concentration.

Self-Driving Cars - Machine Learning

Needs to work for 100% of Cases

Part of getting code to work is that you need to chase down those rare, difficult cases as well.

Below is a difficult case for the self-driving logic, although all the people in the audience understands what they are seeing easily.

alt: bike attached to back of car

Where is the Magic in CS?


alt: ghost input image with foot in the way

Where is the Insight? The Power?


Where is the power in this story?

Fare Well Python Programmers!

In closing, I'll say that teaching this class is very satisfying endeavor - it's great to see the light in someone's eyes when the power we know in CS starts working for that student.

Best of luck with your future projects!