Mushro0m - Invidious

Mushro0m

I do my best to upload consistently. I make remixes, original music, and covers. I am an aspiring musical artist.

Bandcamp: mushr00m.bandcamp.com/
soundcloud: soundcloud.com/mushro0m-465589378
youtube: youtube.com/@CG64Mushro0m

(i usually upload my music on YouTube first before on Bandcamp or other places)

View channel on YouTube

Switch Invidious Instance

Videos

Shorts

Livestreams

Releases

Playlists

Posts

Mushro0m

I don't ever really talk about AI on this channel aside from like the occasional AI slop video because I'm interested in its capabilities, but I want to comment on Open AI's new o3 model because its quite interesting.

Open AI demoed their new AI model yesterday on a livestream, and showed the benchmarks, and it completely crushed all of them, I'm talking like 85% at the bare minimum. Their AI model even crushed the ARC-AGI benchmark with 87.5% on performance, higher than even the human benchmark on performance, which is 85%.

some people see this and automatically say "this is AGI!" but I don't think that makes sense. although I agree that there are still a lot of easy things it fails on, (not to undermine the achievement of getting this high on the ARC-AGI benchmark) there are things that I think are important to consider when evaluating an AI that potentially is generally intelligent:

Does it have general knowledge about the world?
(physics, logic, history, science, etc.)

is it able to recursively self improve?
(by self improve I mean provide output that would enhance the models capabilities, such as better algorithms, better math, better ways to train, etc.)

can it be at least meet humans, if not beyond on benchmarks of math, science, logic, etc. and is not just memorization?

can it be creative, come up with new novel ideas?
(such as writing new stories, new ways to go about doing tasks, etc.)

can it be put inside a physical body and in a timely manner figure out how to control and pilot the robot, and do various physical tasks with it?
(washing dishes taking out trash, doing laundry, putting together furniture, using tools.)

I think these are important things to consider when figuring out if an AI system is general. though one more thing, I think its important to note that if the AI system is able to recursively self improve itself, it might self solve these other problems.

if o3 is able to recursively self improve though, I think that is a reasonable threshold into considering it AGI.

edit: im not really saying that i know for certain o3 is or isnt agi, what im trying to get at is that ill believe its AGI after its able to do recursive self improvement.