Claude Fable 5.2 Just Changed Everything

This was a great AI update! Thanks!

Nothing worse then another AI generated hype video about how good AI is

2:40 It's not cheaper. It runs for a super long time, internally reasoning for ages, spits out some results, which are a lot larger etc – the end result, per task, lands at 33%-100% more expensive than Opus 5.5. That's on high reasoning Fable 5.2. If you even look at the higher reasonings, it'll just eat the entire credit you have. Sure, the results are nicer but… om nom nom nom…

Who was the 5th man to walk on the moon? What was the 5th LLM release to be the best?

This is truly bonkers!

Needs to be used to make Elderscrolls 6 faster.

😂 it's a fucking monster there building 😂

It did this and this …..and on and on…..

Fable 5.2 writes some of the worst copy I've seen all year! It seems too busy trying to complete tasks, optimised for completion it has no concept of nuance. I think we're reaching peak generalisation soon where model's will need to be trained of narrow functions.

Confident nonsense ! 😂😅😊

just wait for next week's best model… Been like that for months

You're getting less disorganized, you must be using Claude :p

Yeah it gives you an ETA usually about 300 to 500% longer than what it really takes 😝😝😝 let's please stop with the stupid AGI talk

TLTW and a bad English. Unwatchable.

Get it to fix Microslop

I prompted mine to do all of these things MONTHS ago….seems like they are really going behind the curve….not very future forward at all….I have had all of my agents doing these things for a long time….if this is new to you, you are WAY behind what AI can be doing…..pretty much, this is a joke….

Can it cure baldness yet?

I am thinking seriously about the future, What to do next for us and our kinds. Which field to choose and stay relevant for long. AI hardware and software combinations are getting more and more efficient and traditional methods of teaching/ learning are getting obsolete. Once, knowledge of some field had been a matter of pride, now has become a set of parameters for LLM and we are looking at the pace of replacement by AI…

2:30 "Smarter, faster and cheaper at the same time and yet Anthropic still says that for everyday tasks you should use Opus. Which tells you exactly what kind of model this is" …
Yes it does. And I would be as excited as you if not for the timing – Anthropic are struggling for money and are just about to go public. And just now they release a "God-tier model" for cheap but they beg you to not use it as a cheap model…Why? Because it bleeds money and they can't afford any of it. I know I may be a little pessimistic but this looks to be just an IPO hype before the launch. I'm not buying any of it.

Is it so difficult to understand the difference between AGI and ASI? every time there is dumb people using AGI loosly. This development might well be considered ASI but not AGI it cannot perform any activity a human could but surpass us (but other models already do) in specific areas.

In mid 2027 the models are potent enough to acclerate every part. So they should slow down there. Work that needs years will be done in couple of weeks/days.

Next TPU will be completly designed by googles AI. Not need to push further, just because AI could.

Google will soon take the lead in the race

Hardly. ChatGPT Astra outdoes Fable 5.2 on careful modeling tasks. Astra gets much more accurate detail on very complex reconstructions of Meshy and Tripo models and the parent 2D images that created those Meshy/Tripo models. Fable 5.2 does great coding on ultra complex and large software engineering, but still requires Astra review to fix its mistakes.

I tried it out yesterday to assist me in developing a whitepaper on Human-Centered AI. I took a bunch of my notes and set it loose on the task, not expecting much. But it was incredibly impressive how well it was able to come up with both some original thinking and do its own research. It pointed out areas, where not enough data existed or where it doesn't know something. It created some strong analogies and created a 40 page document that was a shockingly good first draft. It was way beyond what I would have gotten from a different model.