Howdy from Durham,

OpenAI dropped their latest GPT model on Tuesday.

No worries if you missed it.

We’ll do a quick recap here.

Remember when Steve Jobs used to drop heat at Apple Demo Days?

You know what I’m talkin’ about.

You’d hear about an upcoming Apple event and consequently prepare for your mind to be blown (and your credit card to be swiped if your phone was on it’s last legs).

AI companies are doing the same kinda presentations these days.

In fact, on Tuesday, Open AI had a 30min demo of their latest large language model (LLM).

I’m going to break down the 3 coolest parts of the demo in the following format:

Input - what Greg Brockman, president of OpenAI, entered into GPT-4

Output - what GPT-4 spit back

Why that’s wild - self-explanatory

Application - how this tech could be used elsewhere

1. Turning a pen/paper mock-up into a website

Input: Greg Brockman uploaded a picture of a paper/pen sketch and asked GPT-4 to turn it into a website. The sketch was a rectangular box titled “My Joke Website” where the user could click to reveal a punchline to a joke.

Output: GPT built a functioning website in HTML that mirrored his pen and paper sketch

Why that’s wild: AI accurately extracted all of the handwritten text (using OCR), interpreted “push to reveal punchline,” and populated the site with it’s own joke (using NLP)

Application: software developers will be able to quickly mock a webpage, app, dashboard, etc.. and quickly spit out a quick draft in no time.

2. Answering a nuanced tax question

Input: Greg Brockman copy/pasted a 16 page tax document and asked GPT-4 to answer a very tricky tax question based upon a clause buried within the document

Output: GPT-4 answers the document with precision.

Why that’s wild: AI didn’t just find the correct answer but provided justification for how it arrived at a specific number by indicating the exact section of the document that it referenced to arrive at the number.

Application: accountants can ask AI a tax question first. Then, once AI tracks down the answer, the accountant can verify the integrity of the answer by reviewing the clause that AI cites.

3. Passing the bar exam

Input: they gave GPT-4 a bar exam.

Output: it passed with flying colors, placing in the 90th percentile.

Why that’s wild: GPT-4’s predecessor, GPT-3, placed in the 10th percentile on the last test. We’re making large advancements with incredible speed.

Application: I am not quite sure how this will be used for good, but I do know that standardized testing professionals must be paying close attention to this.

What I’m paying attention to:

The major AI competitors are going punch for punch right now with Google announcing their own host of features this past week.

Photo of the week:

Thanks for reading

Which of these 3 things blew your mind the most?

Reply and let me know!

Josh

Want more Build content? Check out the links below

Previous Newsletters // Essays // Twitter // LinkedIn