I like the FX predictions - as someone not in the industry, I'm guessing there's ton of stuff in the works that would blow us all away. I believe one of the Russo Brothers said that he expects fully AI-created movies in about 3 years.
Meta releases Llama 3 - to me this feels like a given. It's clear that Meta is very much betting on being the champion of open-source, so we might even see more than a single iteration next year.
Apple's AI - you're in good company here, as most observers and experts are fully expecting Apple to pull the trigger on AI given their ecosystem and experience. This article from September claims Apple's spending "millions of dollars a day" to develop AI (https://www.theinformation.com/articles/apple-boosts-spending-to-develop-conversational-ai). I'd be shocked if they had nothing to show for it in 2024.
A prominent AI company goes bust. I'd like to add even more specific speculations here:
- Stability AI is a solid candidate for some "founder/legal drama," as you put it. There have already been rumors and controversy swirling about Emad's approach and prominent employees leaving over dissatisfaction with how he runs Stability AI. Midjourney could also get into more legal trouble, since people have found that V6 can replicate movie scenes shot-for-shot with very simple prompts.
-"Flashy piece of AI-enabled hardware is a complete flop." I think we already have that, and it's called Humane AI Pin. I just don't see how this thing has any future in a world where Google Glass failed while actually being a more reasonable product.
A realistic AI-generated video goes viral: Not only that. My bets are on complete movies from a single text prompt becoming a reality. Perhaps not feature-length in 2024, but at least short ones. Text-to-video is progressing very rapidly.
A major election is derailed by AI: There's no doubt that, as with any election, there'll be disinformation efforts. And it'd be naive to assume that ill-intentioned actors won't make use of AI now that they have access to it. The only question is how we define "derailed." With most elections, it's very hard to pin the outcome on one specific factor. Russia's efforts to influence the 2016 US election are extensive and well-documented, but I don't believe anyone's ever made a call about them conclusively affecting the outcome.
Here are a few of my own to throw in the mix:
- Autonomous agents that actually work - someone will find a way to work around the LLM-hallucination-induced limitations and we'll see truly autonomous agents that can independently complete goals within at least certain narrowly defined disciplines.
- The rise of multi-expert models (ala Mixtral). It appears that the architecture that combines multiple expert models may be the path forward that takes us beyond the current paradigm. (In fact, I recall reading about credible signs pointing to GPT-4 actually being a mixture of 8 expert models rather than a single one.)
Fun exercise.
Let's see how things actually pan out by the end of 2024!
(Though I'm more excited about the unexpected developments we've definitely overlooked than the ones we got wrong from the above list.)
You nailed it - I definitely had Humane AI and Stability AI in the back of my mind, but there's always room for other startups to flop, too!
Regarding the definition of "derailed" - I think AI is going to influence elections in subtle and ultimately plausibly deniable ways. But, I think there will be at least one election in which AI is (rightfully or not) painted as the villain, to the point where many believe it's the reason for the election's outcome. The use your 2016 analogy - Facebook/Cambridge Analytica had an effect, but I believe the backlash was disproportionate to its actual impact. And yet, it was a watershed moment for the general zeitgeist turning against Facebook specifically, and Big Tec in general.
I'd love to read a full-length post with some of your predictions!
Oh yeah, I think AI is very likely to get disproportionate amount of attention in news coverage about elections and other disinformation campaigns. Seems like the next frontier of AI pushback, and probably deservedly so. We'll see some back and forth before we reach a healthy balance of regulation vs. progress here.
Not sure if I'll do a full-fledged predictions post, but I may sneak a few into an article.
Data laundering will accelerate and scale in 2024. Google, Amazon, Meta, etc will increase creating or funding academia/university partnerships and/or non profits for collecting data and training models to skirt copyright/IP issues. See fair use definition.
Charlie, this is great! I agree - it's very easy to come up with a lot of predictions, but this took time and thought to flesh out, and it does not go unappreciated here.
I think a lot of these will come true for sure, but it's really tough to say how individual companies will do in this landscape. Suffice it to say, 2024 is going to be very interesting! And I'm here for it.
I forget where I first heard this, but it is very accurate: things are called AI until they work. For example, Netflix's recommendation engine is artificial intelligence, "smart lending" uses AI, Siri and Alexa and Google Search are AI, and Tesla's automated driving is artificial intelligence. But we refer to them as products instead of merely referencing "AI."
We will see a lot more content and services generated by artificial intelligence without it being called out explicitly as the next new thing. For example, for at least 10 years, sports games have been reported on by artificial intelligence. A scorekeeping app called GameChanger would write a few paragraphs of a new story for a Little League baseball game instantly after the game was over. If your scorekeeping was accurate, it was very good writing. I noticed the pattern in espn.com and other places. Most new stories produce paragraphs of text from data in this way today. It is just not shown off as being something from the latest LLMs.
That's a really great point - the goal posts tend to move so seamlessly that forget what they used to look like. I often think about how in our current AI-obsessed age, we don't even talk about the Turing Test anymore, it's been so thoroughly obsoleted.
These are very plausible. The backlash is definitely coming. I suspect that labour unions will get more involved. More and more people with unregulated white-collar jobs will start to see the writing on the wall and pre-emptively unionise / take industrial action.
I have a sneaking suspicion you're right. The Hollywood writers/actors strikes have in some ways provided a blueprint for other labor unions that want to proactively push back against automation.
The New York Times market cap is just under $8 billion. Not a big deal for these companies. The article about the suit suggests that their data is a big chunk of the OpenAI training data.
Right, Twitter was a personal purchase as well. If Bezos felt like it, he could lease his Washington Post data to Amazon (or anyone else for that matter) for training.
The major downside of purchasing a publication is that it gets you involved in politics whether you want to or not. Not just domestic but global politics.
Have you come across any good examples of synthetic data?
I'm going with two:
1) Movie special FX increasingly using AI where at the end of the year, 90% of a high-budget movie uses AI.
2) A movement of on-device AI where people can truly use "smart" phones or watches
I like the FX predictions - as someone not in the industry, I'm guessing there's ton of stuff in the works that would blow us all away. I believe one of the Russo Brothers said that he expects fully AI-created movies in about 3 years.
I'm easily on board with most of this list!
Here's what I have to add to some of your points:
Meta releases Llama 3 - to me this feels like a given. It's clear that Meta is very much betting on being the champion of open-source, so we might even see more than a single iteration next year.
Apple's AI - you're in good company here, as most observers and experts are fully expecting Apple to pull the trigger on AI given their ecosystem and experience. This article from September claims Apple's spending "millions of dollars a day" to develop AI (https://www.theinformation.com/articles/apple-boosts-spending-to-develop-conversational-ai). I'd be shocked if they had nothing to show for it in 2024.
A prominent AI company goes bust. I'd like to add even more specific speculations here:
- Stability AI is a solid candidate for some "founder/legal drama," as you put it. There have already been rumors and controversy swirling about Emad's approach and prominent employees leaving over dissatisfaction with how he runs Stability AI. Midjourney could also get into more legal trouble, since people have found that V6 can replicate movie scenes shot-for-shot with very simple prompts.
-"Flashy piece of AI-enabled hardware is a complete flop." I think we already have that, and it's called Humane AI Pin. I just don't see how this thing has any future in a world where Google Glass failed while actually being a more reasonable product.
A realistic AI-generated video goes viral: Not only that. My bets are on complete movies from a single text prompt becoming a reality. Perhaps not feature-length in 2024, but at least short ones. Text-to-video is progressing very rapidly.
A major election is derailed by AI: There's no doubt that, as with any election, there'll be disinformation efforts. And it'd be naive to assume that ill-intentioned actors won't make use of AI now that they have access to it. The only question is how we define "derailed." With most elections, it's very hard to pin the outcome on one specific factor. Russia's efforts to influence the 2016 US election are extensive and well-documented, but I don't believe anyone's ever made a call about them conclusively affecting the outcome.
Here are a few of my own to throw in the mix:
- Autonomous agents that actually work - someone will find a way to work around the LLM-hallucination-induced limitations and we'll see truly autonomous agents that can independently complete goals within at least certain narrowly defined disciplines.
- The rise of multi-expert models (ala Mixtral). It appears that the architecture that combines multiple expert models may be the path forward that takes us beyond the current paradigm. (In fact, I recall reading about credible signs pointing to GPT-4 actually being a mixture of 8 expert models rather than a single one.)
Fun exercise.
Let's see how things actually pan out by the end of 2024!
(Though I'm more excited about the unexpected developments we've definitely overlooked than the ones we got wrong from the above list.)
You nailed it - I definitely had Humane AI and Stability AI in the back of my mind, but there's always room for other startups to flop, too!
Regarding the definition of "derailed" - I think AI is going to influence elections in subtle and ultimately plausibly deniable ways. But, I think there will be at least one election in which AI is (rightfully or not) painted as the villain, to the point where many believe it's the reason for the election's outcome. The use your 2016 analogy - Facebook/Cambridge Analytica had an effect, but I believe the backlash was disproportionate to its actual impact. And yet, it was a watershed moment for the general zeitgeist turning against Facebook specifically, and Big Tec in general.
I'd love to read a full-length post with some of your predictions!
Oh yeah, I think AI is very likely to get disproportionate amount of attention in news coverage about elections and other disinformation campaigns. Seems like the next frontier of AI pushback, and probably deservedly so. We'll see some back and forth before we reach a healthy balance of regulation vs. progress here.
Not sure if I'll do a full-fledged predictions post, but I may sneak a few into an article.
Happy New Year!
Data laundering will accelerate and scale in 2024. Google, Amazon, Meta, etc will increase creating or funding academia/university partnerships and/or non profits for collecting data and training models to skirt copyright/IP issues. See fair use definition.
Charlie, this is great! I agree - it's very easy to come up with a lot of predictions, but this took time and thought to flesh out, and it does not go unappreciated here.
I think a lot of these will come true for sure, but it's really tough to say how individual companies will do in this landscape. Suffice it to say, 2024 is going to be very interesting! And I'm here for it.
I forget where I first heard this, but it is very accurate: things are called AI until they work. For example, Netflix's recommendation engine is artificial intelligence, "smart lending" uses AI, Siri and Alexa and Google Search are AI, and Tesla's automated driving is artificial intelligence. But we refer to them as products instead of merely referencing "AI."
We will see a lot more content and services generated by artificial intelligence without it being called out explicitly as the next new thing. For example, for at least 10 years, sports games have been reported on by artificial intelligence. A scorekeeping app called GameChanger would write a few paragraphs of a new story for a Little League baseball game instantly after the game was over. If your scorekeeping was accurate, it was very good writing. I noticed the pattern in espn.com and other places. Most new stories produce paragraphs of text from data in this way today. It is just not shown off as being something from the latest LLMs.
That's a really great point - the goal posts tend to move so seamlessly that forget what they used to look like. I often think about how in our current AI-obsessed age, we don't even talk about the Turing Test anymore, it's been so thoroughly obsoleted.
I suggest you add probability estimates to your predictions!
Good idea! I am way way more confident about some of these than others.
These are very plausible. The backlash is definitely coming. I suspect that labour unions will get more involved. More and more people with unregulated white-collar jobs will start to see the writing on the wall and pre-emptively unionise / take industrial action.
I have a sneaking suspicion you're right. The Hollywood writers/actors strikes have in some ways provided a blueprint for other labor unions that want to proactively push back against automation.
The New York Times market cap is just under $8 billion. Not a big deal for these companies. The article about the suit suggests that their data is a big chunk of the OpenAI training data.
Given that Elon aquired twitter, I am surprised that no one leading ai business has aquired a media publication.
Openai has signed an agreement to incorporating articles of a media publication in chatgpt but that's not the same as embedding.
Jeff Bezos bought the Washington Post about 10 years ago and still owns it.
True, but that was a personal purchase I believe, and separate from Amazon.
Right, Twitter was a personal purchase as well. If Bezos felt like it, he could lease his Washington Post data to Amazon (or anyone else for that matter) for training.
Yeah, I think startups like OpenAI aren’t yet big enough to outright acquire a media company, but Google/Amazon/Microsoft/Meta absolutely are.
The major downside of purchasing a publication is that it gets you involved in politics whether you want to or not. Not just domestic but global politics.
Have you come across any good examples of synthetic data?
That's an interesting line of thought! I hadn't considered that adding more to the "idea space" might result in worse information.