Azyle

We built god-like machines, only to realize we can't scale our way out of being human. Conceive → Create → Become

[Azyle @AzyleTheCreator]

Let me explain why Fable 5 is so special. For a long time, models have been great doers but you had to be the director and idea manufacturer. This stayed true even for models like Opus 4.8 and GPT 5.5 Pro. They ALWAYS converged around the same spectrum of generic opinions.

Fable 5 is the first model where it can be the thinker alongside you. It is the first model that can generate opinions that truly break out of the generic box. Truly truly special.

Lower level people can now have proper guidance.

Higher level people can now have a proper advisor.

The highest level people can now have a quality 2nd set of eyes.

[Azyle @AzyleTheCreator]

@kimmonismus But that's also just matching benchmarks, it still doesn't mean it will feel the same

[Azyle @AzyleTheCreator]

@Adidotdev @thsottiaux @ajambrosino This is very nice dude.

[Azyle @AzyleTheCreator]

@grandpasconcern Far from incompetent. Anything you dislike, was designed to be that way to benefit them.

[Azyle @AzyleTheCreator]

What a tough one. I honestly don't know what's the better path.

[Azyle @AzyleTheCreator]

Dunno. I feel some sort of major architectural change is needed for that, so 16 GB Vram today might not be the version that runs those types of models. It would be a completely different type of 'Vram'.

Coz think about it, a GPU was created for the purpose of gaming, and it happened to be good at AI.

At some point, we should have some sort of component built for AI from the ground up.

So old gen gpu tech runs old gen models.

New gen 'gpu' (Who knows what component it would be) would run new gen models.

Plus we don't know if that kind of tech is gonna price out retail or simply require special licenses.

But regardless, it's hard to see a world where retail gets that sort of access so easily.

[Azyle @AzyleTheCreator]

But at this point in time. Why do we assume impressive SVG is a by product of intelligent/creative thinking instead of just assuming that labs likely now train the models on SVG.

I feel SVG demos are common to the point where labs will try to bench max it, especially if they are already behind on real evaluations.

[Azyle @AzyleTheCreator]

Also we shouldn't be locked into the idea of thinking 'vram' forever - thats just a unit of measurement we are locked into TODAY. Never know where tech can go within the next 5 years.

Especially with the pace of AI development. There's certain creations / objects / processes that one simply can not imagine or fathom until it actually comes into existence.

[Azyle @AzyleTheCreator]

Dunno. I feel some sort of major architectural change is needed for that, so 16 GB Vram today might not be the version that runs those types of models. It would be a completely different type of 'Vram'.

Coz think about it, a GPU was created for the purpose of gaming, and it happened to be good at AI.

At some point, we should have some sort of component built for AI from the ground up.

So old gen gpu tech runs old gen models.

New gen 'gpu' (Who knows what component it would be) would run new gen models.

Plus we don't know if that kind of tech is gonna price out retail or simply require special licenses.

But regardless, it's hard to see a world where retail gets that sort of access so easily.

[Azyle @AzyleTheCreator]

Dunno. I feel some sort of major architectural change is needed for that, so 16 GB Vram today might not be the version that runs those types of models. It would be a completely different type of 'Vram'.

Coz think about it, a GPU was created for the purpose of gaming, and it happened to be good at AI.

At some point, we should have some sort of component built for AI from the ground up.

So old gen gpu tech runs old gen models.

New gen 'gpu' (Who knows what component it would be) would run new gen models.

Plus we don't know if that kind of tech is gonna price out retail or simply require special licenses.

But regardless, it's hard to see a world where retail gets that sort of access so easily.

[VraserX e/acc @VraserX]

I keep wondering when local AI models running on normal 16 GB RAM machines will catch up to what GPT-5.6 Sol and Fable 5 can do today.

My guess is around 18 months. What do you think?

[Azyle @AzyleTheCreator]

@ggg78g89 This is the definition of vibe coded UI. I'm sure a better prompt would make it 100x better though.

[Azyle @AzyleTheCreator]

Will anthropic try to butter us up with opus 5 soon...

[Azyle @AzyleTheCreator]

@StatsWire @synthwavedd Doesn't make sense given that they're Google though.

[Azyle @AzyleTheCreator]

@synthwavedd Even if true, useless imo.

[Azyle @AzyleTheCreator]

@localghost Weird. I see a lot of people have the exact opposite experience. I wonder which is true.

[Azyle @AzyleTheCreator]

@jp54362 Coz if chatgpt does it, then claude will have to do it too, and they both end up making less money.

[Azyle @AzyleTheCreator]

@saltjsx But what can it do that chatgpt or claude can't do. I think you should demo a little something about that. Even if it's just a better UI or personality.

[Azyle @AzyleTheCreator]

@beflowq @LexnLin Probably anything truly useful, do what you feel you can and let's see what people say

[Azyle @AzyleTheCreator]

@Ananth7e @thsottiaux @ajambrosino Yes, would be very helpful

[Azyle @AzyleTheCreator]

@iambokensha @argofowl True. I once used 3 max 20x plans alongside 1 pro 20x plan. Would be dope if claude at least offered unlimited core chat as well.

[Azyle @AzyleTheCreator]

@iambokensha @argofowl Maybe I will join you soon. But then again, wanna wait and see if I need multiple chatgpt pro plans when sol comes out.

[Azyle @AzyleTheCreator]

Funny thing I noticed. A lot of you would get way better output from an employee instead of a massive token spend...

Employees & AI isn't always the same thing.

I was a victim of this mentality by the way.

[Azyle @AzyleTheCreator]

@ant_dracula What models do you use for these tasks