ChatGPT: "A New Technology Adjusts Your Thinking About Computing"

Do you know what ChatGPT is? There's a good chance that the answer is "no" because, last time I looked, mainstream media like the New York Times or the Washington Post have had zero articles about it. But on Twitter, over the last week or two, it has been about 10% of the posts in my feed, I would guess. It went from zero users to over a million in five days (compared to at least a year for Google, YouTube, or Facebook etc).

As economist Tyler Cowen said in his Marginal Revolution blog:

More than any other time, if you are not on Twitter, you just don’t know what is going on.

The quote I used in the title is just a few words of this comment by Box CEO Aaron Levie. Here's the full quote:

There’s a certain feeling that happens when a new technology adjusts your thinking about computing. Google did it. Firefox did it. AWS did it. iPhone did it. OpenAI is doing it with ChatGPT.

So what is ChatGPT? ChatGPT is a development by OpenAI, with CEO Sam Altman and backing from Microsoft, Elon Musk, and others. This is the page where you sign up or log in. This is Sam Altman's tweet announcing availability (scan the whole thread). As I already said, five days after this tweet, there were over a million users. You basically tell it what you want, and it does its best to give it to you.

Or in OpenAI's own words:

We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.

How big is the model? I don't think there is an official number, but since it is based on GPT-3.5 the answer is probably about 175B parameters. Yes, 175 billion. In case it's not obvious, that is a huge number. But we do huge numbers in semiconductors. Assuming it is using NVIDIA  A100 GPUs (the H100 is not yet available), those parameters fit into just 5 80GB A100 GPUs, so fits on an 8-GPU server (available on Azure Cloud).

Examples

The best way to see what OpenAI is like is through examples. I encourage you to use Google to find some since I can only put a couple here. Note that the input is the first line, and all the rest was generated automatically by ChatGPT:

Why is Spirited Away such a popular film?

Spirited Away is a highly acclaimed and widely beloved animated film from Japan. It was written and directed by the renowned Japanese filmmaker Hayao Miyazaki, and was produced by the studio he co-founded, Studio Ghibli. The film tells the story of a young girl named Chihiro who becomes trapped in a strange and magical world and must find a way to escape and return home.

Spirited Away was a huge commercial success, becoming the highest-grossing film in Japanese history at the time of its release. It also received widespread critical acclaim and won numerous awards, including the Academy Award for Best Animated Feature.

One of the reasons that Spirited Away is such a popular film is its beautiful and imaginative animation, which is filled with fantastical creatures and landscapes. The film also has a strong and relatable protagonist in Chihiro, and a compelling story that is both thrilling and emotionally resonant. Additionally, the film's themes of loss, growing up, and the power of friendship and family have resonated with audiences around the world. Overall, the combination of stunning visuals, compelling storytelling, and universal themes have made Spirited Away a beloved and enduring classic of animated cinema.

Or how about debugging some code. Again, the query is the first few lines, and all the rest (starting at "The code contains a bug") was generated by ChatGPT:

Find the bug with this code:
for (var i = 0; i < 5; it+) {
setTimeout(() => console.log(i), 1000)
}

The code contains a bug that will cause it to always print the number 5 to the console, rather than the expected sequence of numbers from O to 4. This is because the setTimeout' function creates a closure that captures the value of the "i' variable, but the "i" variable is changed before the callback function is executed. To fix this bug, the code can be modified to create a new variable in the loop that is passed to the 'setTimeout' callback function, rather than using the loop variable directly. This will ensure that each callback function captures the correct value of the loop variable, and the expected output will be produced. Here is an example of how the code could be fixed:
for (var i = 0; i < 5; i++) {
// Create a new variable and pass its value to setTimeout
const value = i;
setTimeout(() => console.log(value), 1000)
}
This code will print the numbers O to 4 to the console, as expected. Alternatively, the "let' keyword can be used in place of "var" to declare the loop variable, which will automatically create a new variable for each iteration of the loop and avoid the need to create a new variable manually. This is a common pattern for avoiding closure-related bugs in JavaScript.

ChatGPT Does Me

write a blog post on the topic of static timing analysis like paul mclellan does on the cadence website

As the complexity of electronic systems continues to increase, the importance of static timing analysis (STA) has become increasingly clear. This technique is used to verify the timing of a digital circuit, ensuring that it will operate correctly and meet its performance requirements.

At its core, STA is a process of evaluating the timing of a circuit by analyzing its design, without the need for actual simulation or testing. This allows designers to identify and fix timing issues early in the design process, before they become costly and time-consuming problems.

One of the key benefits of STA is its ability to save time and resources. By identifying and fixing timing issues early in the design process, STA helps designers avoid the need for expensive and time-consuming iterations of simulation and testing. This can save significant time and effort, and help to accelerate the overall design process.

Another benefit of STA is its ability to improve the reliability and performance of digital circuits. By ensuring that a circuit will operate correctly and meet its performance requirements, STA helps to reduce the risk of errors and defects, and can improve the overall quality of the final product.

At Cadence Design Systems, we are committed to helping our customers optimize their design processes and achieve the best possible performance and reliability. Our industry-leading STA solutions are designed to help designers quickly and accurately verify the timing of their digital circuits, and identify and fix any potential issues.

OK, there are bits that are not 100% accurate (or maybe don't make 100% sense). Plus, I never write press-release-speak stuff like the last paragraph. But it is disturbingly good.

Conclusion

I think that this is a big deal and that you owe it to yourself to take a look.

Let's finish with a bit of fun:

 

Sign up for Sunday Brunch, the weekly Breakfast Bytes email.

.