A couple of weeks ago I published an article on Medium entitled: “Top 5 AI Predictions for 2023.” You probably didn’t read it because I didn’t publish it here. But don’t worry, you didn’t miss anything: It was an experiment I did with ChatGPT. As I expected, the predictions it came up with were plausible but generic and boring.
Today I’ll share with you my true top AI predictions for 2023.
The listicle below is an extension of the arguments I’ve been laying out in TAB throughout the year, projected into the future. As you can expect, most of my predictions are on generative AI, which will remain a strong focus of the AI conversation. At the end, there’s a bonus section on “things that won’t happen in 2023.”
I find it extremely hard to make precise and accurate predictions. A single event—like OpenAI’s 2020 paper on the scaling laws for language models (LMs), which led to the creation of GPT-3—can radically change the trajectory of the whole field. Also, it’s good to be careful with what we foresee.
But I want to play this game and see how well I do. If you want, too, leave a comment with your hottest predictions for 2023. What's something you think will surely happen and something that you think will happen that most people would disagree with?
At the end of 2023, we’ll see how well we did. I’ll evaluate our predictions against the outcomes (although quantifying our hit-miss ratio is hard, a qualitative evaluation is still a nice exercise).
As a side note: I won’t be publishing articles during the last week of the year. I won’t have much time to write (and you probably won’t have much time to read). Instead, I’ll send you a couple of open threads so you can give me suggestions and ideas to help me adapt TAB as much as possible to your interests in 2023.
My last long-form article of the year will be next Friday (free)—a review of the year at TAB and a summary of my long-form posts (the best resource for a quick glance at the events of 2022). I’ll publish my first long-form article of 2023 on January 3rd (also free). That’s why this one is exclusive to paid subscribers.
The Algorithmic Bridge is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
1. GPT-4 (vs ChatGPT)
As rumors suggest, OpenAI will release GPT-4 in 2023, but it’s possible they won’t do it as soon as we expect.
Demand for ChatGPT was much higher than OpenAI anticipated. The enthusiastic response from users forced the company to increase server capacity a few times. It’s possible that part of the resources they put on ChatGPT were intended for GPT-4.
Another possibility is that GPT-4 is fully trained already. However, after such a success with ChatGPT, it wouldn't make business sense to release GPT-4 too soon. It'd be better for them to wait until people calmed down. They could come back a few months later and make a big entrance—and we know OpenAI likes big entrances.
Whether the rumors are true or not, GPT-4 will be much better than ChatGPT performance-wise. However, it won't be anywhere close to the relative success it should attain—compared to ChatGPT in terms of usage and user reach—if we assumed performance was the driver for user satisfaction.
To put it bluntly: If GPT-4 is, let's say, 10x better than ChatGPT, it won't be anywhere close to being 10x the sensation that ChatGPT has been.
I'd even dare to say—and here’s my hottest take—that GPT-4 will be less of a sensation than ChatGPT. I don't think performance is the main driver for user satisfaction or interest. Here are the three key reasons why I think this will be the case:
Keep reading with a 7-day free trial
Subscribe to The Algorithmic Bridge to keep reading this post and get 7 days of free access to the full post archives.