Anthropic has just DROPPED A NUKE! (Fable 5.2)（原声文稿）

Transcript of Anthropic扔下王炸：Fable 5.2发布 （中文完整文稿）。

Ladies and gentlemen, Fable 5 .2 is here and holy crap is this model good.

In this video I'm going to walk you through it and give you everything you need to know, not just the sexy looking benchmarks, but I'm also going to run it on a bunch of tests and show you guys what it looks like if you have lots of money to throw at it.

So first things first, if you guys haven't checked out the landing page for Fable 5 .2 and Mythos 2, you should. Do you guys see this like sexy little animation as I move my investor around the place? Pretty dang cool, huh?

Worse noting that we're releasing both Fable and Mythos today, but us puny underclassmen only get access to the former.

I'm not gonna be talking about Mythos as much, but obviously it is there and square for people that have cybersecurity backgrounds.

The big thing I'd like to point out that I don't believe other people probably will is Fable 5 .2 is not insane or changes everything or oh my God, the sky is falling down, AGI is here, any more so than previous models.

All Fable 5 .2 is, really, is it's a non -insignificant bump up in capabilities across the board from the current state of the art, Opus 5 .5.

And so I just want you all to keep that in mind as you look at this benchmark review and then later demo test, but also others as well.

Now there's only so much that we could actually show with benchmarks to be clear.

I'm going to just dive right into some demos later, but it is important to understand the relative magnitude of improvement.

And as you guys can see, Opus 5 .5 versus Fable 5 .2 has about a 5 to 15 % difference across the board.

And so Opus 5 .5 on Agentec coatings Frontier Code 2 .0 scored 45 .5, Fable 5 .2 scored 49 .1. Do the math, that's about 10%.

Agentec coating Frontier SWV2 was 62 .3 versus Fable 5 .2, 69 .9.

Again, if you just look at it as a relative portion, that's 10%.

Okay. And so, agentic scientific research is a little bit bigger of a bump.

It's 15 to 20 percent. Business workflows, 15 to 20 percent.

But I basically just want you to think of Fable 5 .2 as, you know, the war -tornal versus, you know, the opus 5 .5 squirtle. It's just a little bit better. It's got some cool badass sunglasses. And it's here to, you know, pardon my French, fuck some shit up.

So, I can talk all day about all of the incredible scientific advancements, the fact that Anthropics using Fable 5 .2 or models like it to merge 90 % of all their code, the fact that it is just crushing it on both cybersecurity and then literally solving open problems in theoretical physics that involve black holes, and then protein folding.

But I think what you guys are probably here for is really to see what it looks like if you throw it unhindered at a bunch of tasks. So let's go and do that, shall we?

First things first, if you just start a new chat with Fable 5 .2 asking it what model it is, Anthropic will pump a little notification up somewhere in the top right -hand corner saying, you guys get a free usage reset.

And so I'd highly recommend get out there and make use of that free usage reset to test it.

Now, the real mark of intelligence nowadays is not necessarily my numbers bigger than your number on a benchmark. To me, it is can the model intuit what it is that you want?

And so for what I'm about to show you, a bunch of demos, I simply ask Fable 5 .2, hey, I'm gonna show this to at least over 100 ,000 people, hopefully that's up to you. to subscribe and like for more.

Can you build me a set of extraordinarily impressive demos that I could show to them in under 10 minutes? And here's what it got back to me with.

This is an index page that shows virtually whatever the heck I want.

And so one demo here is type anything. Fable 5 .2 is legitimately just introducing itself to you here. And I'm not kidding. I mean, I actually have played this back three or four times now. It's actually saying, hey, I'm Fable 5 .2, you're 100 ,000 people.

you know, I built this in 10 minutes.

By the way, if you want to try typing stuff, feel free. And so I am gonna type hello.

And you guys could see here, we just put together probably one of the sexiest typing demos ever, right?

Like we could do anything we want here. I mean, this is so snappy, it's not even funny. We have like deformation, I can click this. It's super optimized and it's very, very dynamic and engaging. This is again just something I didn't have to ask it to do.

I didn't have to say make a cool typography demo.

It just came up with all this shit itself as part of the demonstration It also made a landing page and this is Volta over here worth noting it made all of this in SVG So this is like coded and created from the ground up every spoke on this dang bike This is very very engaging and very high quality I'm a former web developer and seeing stuff like this is both really exciting and also kind of depressing because back in the day A website like this would have taken me legitimately weeks or months I mean, just to make the bike probably would have taken me a week, and now I can actually modify it as I want and, you know, change metrics.

Worth noting there's another landing page here, too.

For whatever landing page you want, better design sense across the board, none of this is tacky. We're well past tacky taste now.

I'd say taste gate is over, and Fable 5 .2 probably has better taste than I do in many respects. You know, your designs are just going to appear significantly more unified.

They're going to be very interesting, but clean and not over the top, you know, visual effects, you know, CTAs and so on and so forth. They're just going to be a lot tighter. And yeah, I'm just opening up these couple of landing pages. You guys can immediately see we're already past replace every web developer on the planet.

I was really interested in simulations here because I think as the models grow more intelligent, simulations essentially become the demonstration of like their cap or their threshold.

And so what I have here appears to be a bunch of chemicals reacting and spreading on my GPU directly.

This is sort of a coral design and obviously I can work through it and I can destroy it and have it reform There's also sort of like a mitosis design where you have a really small initial set of particles and then these start multiplying and growing There's also like a maze design.

There's like a spot design, which is it basically duplicating I mean this one in particular is kind of cool But also fairly terrifying because this is what's going on in our bodies 24 -7 got some very high -quality cloth simulations This is not super special or different compared to what Opus 5 .5 could do.

Beautiful fireworks in here. Again, just took me a couple of minutes to set up.

I could also, you know, fire these or make things happen anytime I want. Although I'm going to turn the sound off there before it makes me deaf.

And you can also fly over an endless generated Fable 5 .2 landscape as you guys could see here.

I want you guys to know that, you know, while this stuff was possible before, the optimization on my GPU and my computer is next level.

this is smooth as hell and it's capable of rendering at far higher levels of speed than I was before. Just look at the quality of that water.

Seeing a lot of high quality particle designs here, this here is a graph where I can actually change the figure of sound on my computer.

I don't know if it's clear guys, but we're at the point where you can do literally whatever you want. Like I mean this entirely seriously.

I am controlling the sun in a virtual world right now and the total sunk cost for me here was probably less than 30 cents to maybe 50 cents in tokens.

You know, I just one shot at this. There's no optimization going on here.

There's absolutely nothing you guys could take any one of the demos that I just showed you and turn them into extraordinarily engaging experiences whole worlds really, I mean, entire universes, like this is a freaking desktop inside of my desktop.

We just built what is essentially like, I don't know, almost an operating system.

I could type, I could do whatever I want, I could list things. I mean, this is absolutely mind boggling.

How about using this artificial paint tool to draw inside of my simulated computer, inside of my real computer?

So why am I so like crazy stoked about this?

Again, this is just a significant improvement on its predecessor.

This is not, you know, making paper clips of humanity or anything like that. At least I certainly hope not Fable 5 .2. And if you do, remember me.

This is probably capable of doing anything that any white a color knowledge worker could do.

I've been using even Opus 5 .5, but now Fable 5 .2, not just as a coding tool anymore. I'm literally using it like a colleague.

I think we're well beyond junior level now, probably squarely into mid -level, higher level.

That's a mouthful. But nothing here is going to stop.

These models are just going to grow more intelligent and more capable of understanding what it is that you really want, not just what you can verbalize.

Eventually, we'll get to the point where these models probably anticipate more of your needs than anybody else, and what the world's going to look like when that happens, I don't know.

Anywho, hopefully you guys appreciated this fabled 5 .2 review.

Stay tuned, I'm going to do a much more in -depth breakdown soon after I have the time to let it rinse my bank account and my wallet.

Subscribe to the channel, and I'll see you there.