Acceleration.

7 days of new AI technologies shows us that everything is happening very fast.

I honestly can’t believe that the pace of generative AI continues to accelerate. Within the past week alone, significant progress has been made in the realm of AI tools - tools which are already accessible to the general public. And, with those tools, we are beginning to see, however blurred, the outlines of how our work and lives might change with AI.

There is a lot to discuss, so, rather than this being a deep post, it will be a shallow, but broad, one. I want to outline what has happened in the last week or so, and, more importantly, what has changed about how we should think about our AI era. Consider it a preview of future discussions.

What happened?

So, first what happened in the last seven days? Two major new AI models were released to the public (one mind-blowing, on disappointing); the world of AI image creation leapt forward; and a whole lot of large companies put out specialized products that would have been absolutely disruptive just a few weeks ago, but which are barely noticeable with the background buzz of accelerating technology.

Lets start with the disappointing model first. That would be Google’s long-awaited Bard. I have had access for 24 hours, but so far it is… not great. It it seems to both hallucinate (make up information) more than other AIs and provide worse initial answers. Take a look at the comparison between Google’s Bard and Microsoft’s Bing AI (based on GPT-4, as we learned last week, more on this in a minute), answering the prompt: Read the PDF of New Modes of Learning Enabled by AI Chatbots: Three Methods and Assignments and perform a draft critique of it. Bard, which Google says is supplemented by searches, gets everything wrong. Bing gives a solid and even thoughtful-feeling critique, and provides sources (though these can be hit or miss, and Bing still hallucinates, just less often)

Again, Bard was just released, so I may not have figured out the secrets of working with it yet. But Bard also fails at generating ideas, at poetry, at helping learn and explain things, at finding interesting connections, etc. I can’t find a use case for this tool yet, it feels incredibly far behind ChatGPT and its competitors. I am not sure why Bard is so mediocre. Google has a lot of talent and many models, so maybe this is just the start.

So lets turn the big competitor - GPT-4, from OpenAI. If you have been using ChatGPT since November, you have been using GPT-3.5. As of last week, paying customers have access to the newest model, GPT-4. It is incredibly impressive. It also turns out that Bing was running on GPT-4 all along¹, so you may want to look at the list of surprising things Bing does in order to get a sense of GPT-4’s capabilities. I will dive a lot more into GPT-4 in the future, but as one example of its power, I have been using it to write programs in Python and Unity (programming languages I literally do not know at all!) by just telling it what I want in words: "I need to create an Amazon Echo skill that will flash my hue lights green and blue when I yell party. Can you create it?" It did, and now my lights flash blue and green.

Also, it appears to pass the Bar, LSAT, SAT, the Medical Licensing Exam…

The third major updated product is Midjourney, the image generation AI. It now generates photo-realistic pictures with a simple sentence. Here’s “a woman climbing a hill towards the camera.” It takes effort to see that it isn’t real.

And this isn’t even the end of what happened this week: Microsoft rolled out a new coding companion (the original, primitive, one already cut programming time in half in one controlled study), learning companies like Duolingo and Khan academy are teaching with ChatGPT tutors, Adobe released AI creative tools, and lots more. Each of these moves are interesting, but, for this post, lets move to the bigger questions:

What does it mean?

Here are a few thoughts on initial implications after experiencing these new tools:

Even if AI technology did not advance past today, it would be enough for transformation. GPT-4 is more than capable of automating, or assisting with, vast amounts of highly-skilled work without any additional upgrades or improvements. Programmers, analysts, marketing writers, and many other jobs will find huge benefits to working with AI, and huge risks if they do not learn to use these tools. We will need to rethink testing and certification, with AI already able to pass many of our most-challenging tests. Education will need to evolve. A lot is going to change, and these tools are still improving.

Some of the risks of highly-capable AIs is becoming clear. OpenAI released a white paper showing how GPT-4 was capable of some dangerous acts, from giving accurate advice on money laundering to writing threatening messages to people, if it wasn’t stopped by the system’s internal guardrails. GPT-4 was even able to develop, and order online, chemicals that it built to matched the properties of known compounds. And better image generation makes fakes easy - you really can’t trust any photo or video anymore. For a lighter take on how easy this is, below is a picture I generated of an iPhone photo of the Loch Ness monster, but it also works for political figures and more. We are not ready for the disruption that AIs, especially those that can create realistic deepfake images and videos, will bring, and those capabilities are already here.

Things are not slowing down. If we had hoped for a breather to absorb all of this new technology, it isn’t going to happen. Too many organizations are incentivized to release AI products to imagine anything slowing down. Likely, between the time I send this email and you read it, Microsoft will have added image generation capabilities to Bing, continuing the blistering pace of AI development. And, while there may be limits to the AI technologies like Large Language Models, we haven’t reached them yet. And there there is this extraordinary new paper by Microsoft researchers, full of examples, in which they conclude, rather mind-blowingly, that GPT-4 “could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system.” I don’t think anyone knows what the next few years of technology will look like.

This wasn’t supposed to be an AI-focused Substack. My original idea was that I would write about a different management paper every post (“One Useful Thing”), not a single topic. And that is what I did until the end of November, when ChatGPT turned me from an AI-skeptic to an AI-believer. I have only come to believe more fervently that the world is about to change in complex ways, both good and bad. And I think the only way to prepare for the future is to get as comfortable as possible with the AIs available today. Everyone should practice using them for work and personal tasks, so that you can understand their uses and limits. Things are only going to accelerate further from here.

1. My guide to the weirdness that is Bing.