2023 certainly ended nothing like it started1. Will 2024 be the same?
How could it not? The tech world is quite turbulent at the moment, especially around AI. But I think this turbulence has a trajectory that makes it predictable.
Here they are, my predictions for AI in 2024.
2024 for AI as a Technology
2024 is the Year of Documents … RAG and related capabilities like Internet Search and Functions.
I discussed this a bit more in my previous post:So I’ll just summarize here:
GenAI has been a bit like a closed book test: whatever answer the model comes up with is based upon what it (mis-)remembers from training. So, sure, George Washington, the founding father, invented the peanut2. But Retrieval Augmented Generation is like an open book test where the underlying models get to consult external resources to help come up with answers. It’s going to do much better.
We’re just on the cusp of people realizing that they just need to upload any sort of knowledge base they have to get a custom ChatBot that can produce well-informed answers to questions. The simplicity of the effort should not hide the power of the results. In 2024 people will become comfortable using AI to build tools for themselves.
That’s really exciting. And scary.The capabilities of generative AI have plateaued for now.
By plateaued, I mean that we won’t see a new model or offering that is an order of magnitude better than what we have today in GPT-4 and similar for several years. New models will be more of a marketing event than a technological breakthrough.
Why?
The powers of a model are related to the size of the model, and the size of the model drives the amount of hardware (especially the hard-to-obtain GPUs) needed to run it. With Moore’s Law sputtering at best, hardware is going to be a (pardon the pun) hard problem to solve. At this point, it’s nearly impossible to even buy enough GPUs for current LLMs, as Oracle admitted when it failed to meet Elon Musk’s xAI requirements.
Then there’s the possible black swan events with TSMC having virtually the only fabs that can build high-end GPUs: an attack by China on Taiwan would bring AI advancements to a nearly dead stop for years.3
That doesn’t mean the very modest, incremental improvements in models won’t be passed off as groundbreaking new capabilities. This is still the software biz.
But … AI applications will take off in 2024.
We just had a revolution in AI a year ago. It takes a wee bit longer than a year to turn a brand new technology into mature products. 2024 will see a bumper crop of innovative products come to market, and maybe by the end of the year we will see AI adoption start to take off Check out Studio AI from a Vlocity alum for a great example of what people are doing.Artificial General Intelligence, AGI, Q*, or whatever you want to call your version of The Terminator, isn’t happening, and certainly not in 2024.
Why? Two reasons.AGI is going to be so much more hardware intensive than current GenAI is, at least two orders of magnitude more. There isn’t the hardware on the market to support building it, nor a market to consume it at a profitable price point (if you could build it).
The state of AI isn’t advanced enough to pull this off. We still have self-driving cars that like to stop randomly in traffic or run into left turning trucks without slowing down4. GenAI still suffers from the Dunning–Kruger effect: it will do something blatantly wrong but with supreme confidence. So if you know someone in the back room of OpenAI who has AGI working, tell them they need to get AGI to fix GPT-4’s hallucinations problem right damn now, please and thank you.
2024 for AI as a Business
Because the underlying AI models won’t get a lot better in 2024, vendors will strengthen their platforms around them to become or remain competitive, and emphasize how “real people” can benefit from AI.
This is already underway: OpenAI released GPTs and Assistants in November. At the same event, they featured a video of various folks saying how they got a benefit out of their personal use of AI.
It’s going to intensify in 2024, and vendors will work hard to make it look like ordinary people are doing incredible things with AI to encourage more people to give it a whirl.Any AI applications designed for the state-of-the-art circa the first half of 2023 will be dead in 2024.
With AI still evolving (even if individual models are relatively stagnant), there’s two directions one can go in to add value. The first is to build valuable and unique business solutions on top of GenAI. The unique value is probably not be in the use of AI itself, but in all the work the product does to incorporate AI into a business process. The second is to build tools that bring state-of-the-art capabilities to a broader market.If you do the first, the evolution of AI offers the chance to improve your business solutions, at your leisure: people want your product because it solves a problem well. If you do the second, you’ve signed up to keep your offering state-of-the-art with constant revisions: people want your product because it makes it easer to use the latest and greatest. But if you do neither, and merely build lightweight apps on top of AI, your products will be out of date before they get out the door.5
And there’s a lot of “do neither” out there.Enterprise spending on AI (especially generative AI) will remain much more modest than we’d all like it to be in 2024, (but will pick up towards the end of the year presaging a great 2025.
There will be several reasons for this. From a macro economics perspective, we’re still in a turbulent economic environment, and investment funding is hard to come by; the ROI of AI investments will be unclear and unproven. Making things worse, next year, in the USA, is a Presidential election, and that creates a lot of uncertainty that causes people to postpone commitments6.The OpenAI fiasco will have no lingering effects and, in fact, OpenAI will strengthen its position as the market leader (with partner Microsoft of course).
Google will continue to try to displace OpenAI but fail.
While Google is trying to create competitive offerings, OpenAI+Microsoft are sewing up the business relationships needed to be the preferred vendor. As the columnist John Dvorak once observed, for a technology to displace a market leader it has to be an order of magnitude better or the inertia and costs of switching aren’t worth it. But because of technology limitations discussed above, Google can’t come up with something that’s an order of magnitude better.7
Nobody else is going to displace OpenAI either, since their cost of capital is going to be higher than Google’s. That doesn’t mean multiple companies won’t come to market with interesting models, and it doesn’t mean people won’t get excited over them. I’m just saying that while the MGB was a fun car8, you can still buy a new Corvette Stingray today.
2024 for AI in Society
US Presidential Election
This is going to be a watershed moment for AI. The amount of deepfakes, AI generated videos, and other AI content is going to be astonishing. We know it’s coming, and we’re still going to be surprised. AI generated content is out there now, but usually carefully presented as parody and humor because, I assume, it’s too soon.9 But the big guns will be brought out in October. Will it work? Of course. Too many people are willing to spend too much money.There will be no significant new AI legislation in the USA. There will be a few meaningless laws passed to try to look like something’s being done, which is basically what the EU has done10.
The election might provide enough of a shock to jump start real regulation, but since the government won’t change until January of 2025, nothing can happen until then. So maybe 2025 will be the year of regulation. Let’s book note that for next year’s predictions.The popular press will continue to misunderstand what’s going on, report on the unimportant, and ignore the important. This is probably not worthy of a prediction, it’s so … predictable!
Final Thoughts
It’s still plenty early for you to stake a claim in the world of AI. Don’t be shy, go build something, anything, and get it out there. The experience will be incredibly valuable even if the product isn’t a hit. Even if you’re not a coder, there’s plenty you can do now. I’m not going to suggest to you that there’s a magic prompt, or magic toolset that you should use. That doesn’t matter. What’s matters are the solutions you come up with!
For me, especially! I started the year as VP for Industry Architecture at Salesforce, and ended it as an AI consultant building predictive and generative AI applications for multiple companies and writing a blog on AI. I did have a bit of a head start, fortunately.
Of course he didn’t. That was Washington Irving. No? How about George Washington Carver?
Yes, capacity is coming on board in other places. But this is a 2024 prediction only.
Also, let’s hope that doesn’t happen, it would be stupid for all involved. Unfortunately, Barbara Tuchman has something depressing to say about countries behaving stupidly.
2 million of them have been called back to the shop for an upgrade!
Since I tend to talk about CRM and Salesforce a lot, I’ll just predict that there’s at least one big pivot coming for Salesforce in 2024. Maybe I should say I hope there’s a big pivot coming. I noticed they just announced support for RAG and vector databases, which is a step into the future.
My wife used to own a retail business, and she noticed the correlation between a drop in business and the run-up to an election. Who cares if correlation isn’t causation if business is down?
… without faked demos.
Fun, but not great for so many reasons.
An article on the topic in The NY Times (free-gifted link): Inside the Troll Army Waging Trump’s Online Campaign