I'd throw Genesis into your roundup of world models (https://genesis-embodied-ai.github.io/). While Sora appears to mimic real-world physics without truly understanding it (much like LLMs mimic language), Genesis seems to be a genuine physics engine that can simulate real-world scenarios from mere text prompts.
But your post anchored for me just how much stuff happened this year alone. Some of these feel like they've been around for well over a year to me, and yet they're not.
Thanks for the links, I'll be checking these out this week. I think world models are incredibly interesting but I've been left wondering whether world models will have the same impact as language models.
Their immediate applications definitely seem more niche, so I don't expect them to have the same measurable mass appeal, but within certain areas of research, robotics, game development, etc. they will probably have a huge impact.
Speaking of mimicking physics, have you (either of you) gone down any Stephen Wolfram rabbit holes? That's kind of his whole thing, and his thinking keeps coming up in different areas for me. I've been reluctantly forced to listen to more of his uncharismatic (but brilliant) ass.
Can't say that I have. I'm only vaguely familiar with him via the WolframAlpha site, which was also one of the very first custom GPTs released by OpenAI (back when custom GPTs were all the rage).
He broke down how LLMs work a couple of years ago, and I found his explanation of neural networks very good.
One of his main things is to try to recreate the laws of the universe with simple mathematical rules, then simulate that in a powerful computer to see if those simple laws actually play out into emergent complexity.
Remember Conway's cellular automata and The Game of Life? Wolfram kind of picked up the torch and ran with it a bit.
Charlie, I love this phrasing: "moving from novelty to necessity."
That really sums up where we are today, and pinpoints what's so notable about this paradigm as opposed to previous ones. How long did the internet take to go from novelty to necessity? Maybe 5 years for those ahead of the curve, but probably more like 15 years for everyone else.
With social media and smartphones, it has been more like 3-5 years from novelty to necessity.
With gen AI, it's down to like 2 years or less. Holy crap. Time is about to compress even more.
thank you for the recap. If you could add number 11, I'd nominate AI regulation. There was a lot of activity at all levels of government, and a general agreement by (most?) foundation model companies that something is needed.
What's interesting about Sora, though, isn't that it can make slick videos - it seems to have a genuine understanding of how objects move and interact — like hand and fingers passing through solid objects
Yes, more impressive, but that was part of the pre-release at the beginning of the year that could have been doctored or the result of 1000's of attempts. It also features freaky hands with elongated fingers and weird artifacts in the background.
Great list!
I'd throw Genesis into your roundup of world models (https://genesis-embodied-ai.github.io/). While Sora appears to mimic real-world physics without truly understanding it (much like LLMs mimic language), Genesis seems to be a genuine physics engine that can simulate real-world scenarios from mere text prompts.
Also Odyssey's Explorer is another candidate, albeit more limited in scope: https://odyssey.systems/introducing-explorer
But your post anchored for me just how much stuff happened this year alone. Some of these feel like they've been around for well over a year to me, and yet they're not.
Thanks for the links, I'll be checking these out this week. I think world models are incredibly interesting but I've been left wondering whether world models will have the same impact as language models.
Their immediate applications definitely seem more niche, so I don't expect them to have the same measurable mass appeal, but within certain areas of research, robotics, game development, etc. they will probably have a huge impact.
Speaking of mimicking physics, have you (either of you) gone down any Stephen Wolfram rabbit holes? That's kind of his whole thing, and his thinking keeps coming up in different areas for me. I've been reluctantly forced to listen to more of his uncharismatic (but brilliant) ass.
Can't say that I have. I'm only vaguely familiar with him via the WolframAlpha site, which was also one of the very first custom GPTs released by OpenAI (back when custom GPTs were all the rage).
He broke down how LLMs work a couple of years ago, and I found his explanation of neural networks very good.
One of his main things is to try to recreate the laws of the universe with simple mathematical rules, then simulate that in a powerful computer to see if those simple laws actually play out into emergent complexity.
Remember Conway's cellular automata and The Game of Life? Wolfram kind of picked up the torch and ran with it a bit.
Glambase is something I would have to throw in too. AI influencers
Great point about AI influencers. Definitely saw some big headlines this year about AI Instagram models.
Great list.
The landscape is shifting so much that it’s hard to keep up. This is a nice summary at the tail end of the year.
Charlie, I love this phrasing: "moving from novelty to necessity."
That really sums up where we are today, and pinpoints what's so notable about this paradigm as opposed to previous ones. How long did the internet take to go from novelty to necessity? Maybe 5 years for those ahead of the curve, but probably more like 15 years for everyone else.
With social media and smartphones, it has been more like 3-5 years from novelty to necessity.
With gen AI, it's down to like 2 years or less. Holy crap. Time is about to compress even more.
thank you for the recap. If you could add number 11, I'd nominate AI regulation. There was a lot of activity at all levels of government, and a general agreement by (most?) foundation model companies that something is needed.
What's interesting about Sora, though, isn't that it can make slick videos - it seems to have a genuine understanding of how objects move and interact — like hand and fingers passing through solid objects
Hmm, the Sora demo's I saw showed a severe lack of physics understanding. It was completely underwhelming.
There were a lot that were pretty wonky. But I think about this video a lot: https://www.reddit.com/r/ChatGPT/comments/1arp68c/prompt_reflections_in_the_window_of_a_train/ - that the model understood there were people inside the train, and rendered their reflections in the darkness, still blows my mind.
Yes, more impressive, but that was part of the pre-release at the beginning of the year that could have been doctored or the result of 1000's of attempts. It also features freaky hands with elongated fingers and weird artifacts in the background.