Your 2024 Predictions thread

[Powers]

AI will have updates that will be such background changes that the general public with not notice them and feel like it's plateaued and the novelty of this past year will drive AI in the same way as VR, and NFTs.

Gonna have to strongly disagree here. I think 2024 will be just as crazy as last year with regards to AI. There might be a Dotcom-style bubble at some point in the future, but I suspect that's at least a few years away.

Extremely likely in fact I would say.

[Yuli Ban]

AI in video game creation becomes tangibly feasible, with GPT-4.5 capable of creating SNES-quality games

Holy shit. That would be phenomenal. Are you assuming simpler genres like sidescrollers, or games as complex as RPGs? There's obviously a difference between Contra III and Super Mario RPG, but honestly even making something on par with a 1991 SNES sidescroller would be mind-blowing.

Assuming 4.5 has near perfect coherence and near human level basic mathematical abilities, an SNES game is usually not more than a few dozen thousand lines of code. The hard part is data assets.
But a 128k context window, assuming that translates to about 80,000 lines of code, would be enough to program Legend of Zelda: A Link to the Past, which is also ~80k lines of code. If you have all the sprites and audio ready to load in, this theoretical 4.5, if it had an output context window of 16k, could recreate the game in under half an hour.


Theoretically, 4-Turbo could too, but its logical deficiencies are too great.

[raklian]

2024 will be a big year for AI governance as startups try to offer solutions that help companies stay in compliance with AI regulations.

[Dr. Casey]

AI in video game creation becomes tangibly feasible, with GPT-4.5 capable of creating SNES-quality games

Holy shit. That would be phenomenal. Are you assuming simpler genres like sidescrollers, or games as complex as RPGs? There's obviously a difference between Contra III and Super Mario RPG, but honestly even making something on par with a 1991 SNES sidescroller would be mind-blowing.

Assuming 4.5 has near perfect coherence and near human level basic mathematical abilities, an SNES game is usually not more than a few dozen thousand lines of code. The hard part is data assets.
But a 128k context window, assuming that translates to about 80,000 lines of code, would be enough to program Legend of Zelda: A Link to the Past, which is also ~80k lines of code. If you have all the sprites and audio ready to load in, this theoretical 4.5, if it had an output context window of 16k, could recreate the game in under half an hour.


Theoretically, 4-Turbo could too, but its logical deficiencies are too great.

Damn. That's insane. And I guess GPT-5 could make the data assets as well.

Is there a difference between improvements in coherence, and hallucination reduction? It sounds like they're related to each other, but I'm guessing they're not the same thing since you listed them separate in your original post?

Honestly, if 4.5 turns out to be as good as you say - almost perfect coherence, drastic reduction in hallucinations, capable of creating SNES-level games, good at basic math, competent at almost any intellectual task - I'll consider 4.5 to be the world's first (limited) AGI. The limitations being those imposed by context windows, as well as being unable to do certain things if it's not fully multi-modal. In terms of output it would be around the competency level and reliability of a person at most tasks. There's no universal consensus on what constitutes AGI (and Proto-AGI has even less of one), but 4.5 sounds like it might reach the "Does it really matter that much anymore? The weaknesses are disappearing and the capabilities are insane" level.

[Yuli Ban]

Updated my predictions to add a single extra point...

Edit: updated to also change my feelings on GPT-5.