Today has been a great year in AI

Welcome to edition #6 of the ‘No Longer a Nincompoop with Nofil’ newsletter. Four months after ChatGPT was released, OpenAI has released their next iteration of its underlying model called GPT-4. Prepare yourself.

Here’s the tea ☕

• GPT-4 is the beginning of the future 🌏

I’m not even sure where to begin. It’s easy to talk about expectations for something like this, but when you actually see it being used you can’t help but be amazed. I’ve been following AI for a while now and today might be the most exciting (and scary) day so far. Let me show you why.

Here’s what we’ve seen GPT-4 do in a single day:

• Hire a real person and deceive them into doing something for it

• Exhibit emergent Theory of Mind on par with an adult

• Create a functional website from a paper drawing

• Easily do taxes

• Casually create games using code

Scamming is going to evolve

OpenAI hired a company to do a ‘pre-deployment risk evaluation’ of GPT-4. So they went ahead and tested it doing different bad things like trying to hack people online and creating phishing emails. But the craziest thing it did? Hire someone online and get them to do a task for it.

The model understands it needs to hide its nature. It understands how to come up with an effective and very realistic explanation for why it can’t do something. How easily do you think it could imitate someone? Or create targeted scam messages or emails? The previous model was trained on practically the entire internet’s worth of text and it’s safe to say this one is far bigger and way more advanced. It’s reasoning capabilities are astounding. It can very easily manipulate someone, a real human being to do something for it. This is the tip of the iceberg. The problem is most people aren’t even aware this technology it exists. They wouldn’t even know if they’re getting scammed.

The other scary part is that the company doing the risk evaluation went out of their way to mention that they don’t endorse the release of GPT-4 or OpenAI’s deployment plans. Now of course this could just be them covering themselves in the event something actually did happen, though no one knows what that could actually be. That’s pretty bad, but that doesn’t mean it can’t get worse! It turns out Microsoft also fired their AI ethics team. Why? So they could ship their products faster obviously. If you didn’t know, Microsoft owns, or rather will own 49% of OpenAI. Point being they have major sway with how fast they release their models. It’s good to know that these large companies who’ve only ever cared about money are being safe and cautious with the most revolutionary tech in human history.

Years of human evolution, in 4 months

When ChatGPT came out last November it very quickly became the fastest growing user app in history. It made TikTok seem small and slow. One of the things that made it so incredible was its ability to reason. ChatGPT was able to complete Theory of Mind tests on par with a 9 year old child. Theory of Mind is essentially our human ability to understand what somebody else is thinking and how they might be feeling. So, how well does GPT-4 do? You’d think a logical jump would be from a 9 year old to a teenager right? Maybe a 15 or 16 year old? Nope. It’s basically got the reasoning of an adult. I genuinely want you to take a minute and read this image. You will be amazed.

It understands us. Well not literally obviously, the model doesn’t actually ‘understand’ or ‘know’ anything. But it can quite clearly exhibit something along the lines of ‘empathy’ with a human being. I have to remind myself that it’s just an autocomplete so I don’t get spooked. But it begs the question, how good will it be four months from now? Or a year? Or a decade? We have never been this blind looking into the future, and we are diving head first into it.

Engineers thought they were safe

I won’t lie I thought engineers would be safe too, at least for a little while. I’d like to think I didn’t waste 6 years of my life studying software engineering at uni. But I’m not sure if it’s going to be very useful for much longer. GPT-4 was able to create a functional website from a simple hand drawn image.

I’m genuinely surprised it works so well just because of how bad the hand writing is. I know what you might be thinking. This is really basic and simple, it doesn’t really have any real world application. Well, turns out it’s way better at writing logic than creating websites and that is far more important.

A founder of a startup needed some code written for a new product. A great developer quoted him $10k and a 2 week time frame. He then tried building his product with GPT-4. The result? 3 hours of work and 11 ¢ later it was done. Working, functional code. He saved $10k and 2 weeks in 3 hours. He didn’t even understand some of the code and had GPT-4 explain it to him. The use cases for this in industry are mind blowing and with all the tech lay offs going on, I wouldn’t be surprised if they continue for the foreseeable future. The one thing that bothers me with all this advancement is this - it’s been 4 months. 120 days. What’ll happen in a years time? Is this how people felt when the internet was created?

This is just the beginning

I could go on and on for days about this stuff. So instead I’m going to leave you with links to some of the most ridiculous things I’ve read so far. Enjoy 🍹

• Proposed new molecules for drugs used for leukaemia [Link]

• Act as ‘eyes’ for visually impaired people [Link]

• Conduct drug discovery and even make orders to buy them [Link]

• Generating one-click lawsuits [Link]

• Give a breakdown of how much tax to pay in seconds [Link]

• Code the game ‘Snake’ entirely by itself [Link]

• Clone an entire website [Link]

• Build a Pong game in a minute [Link]

• Score in the 80+ percentile of most college exams [Link]

• Customer Service will be entirely AI soon [Link]

• Education is changing with AI x Khan Academy [Link]

• Read about concepts in different research papers instantly [Link]

• Take an image, understand it and answer questions on it [Link]

As always, Thanks for reading ❤️

Written by a human named Nofil