Ancestors

Toot

Written by uis on 2024-12-26 at 23:55

I lately came to conclusion, that what we see as publically avaliable AI models is barely tip of ~~an alicorn.~~ I mean, iceberg.

=> More informations about this toot | More toots from uis@pone.social

Descendants

Written by Cesium on 2024-12-27 at 00:01

@uis did an AI write this joke?

=> More informations about this toot | More toots from inertia@pone.social

Written by uis on 2024-12-27 at 00:28

@inertia does AI passes Turing test now?

No, I was just reading papers 2412.16533v1 and 2402.01680v2. It seems that most advanced publically avaliable AI with even non-free(in both meanings) access does not look as advanced as ones shown in research.

=> More informations about this toot | More toots from uis@pone.social

Written by Cesium on 2024-12-27 at 00:31

@uis I'm pressing x on that. Unverifiable claim that conveniently aids the companies "AGI is only 2 months away fr fr" narrative.

=> More informations about this toot | More toots from inertia@pone.social

Written by Cesium on 2024-12-27 at 00:32

@uis also AI did pass Turing test already, right? Did it?

Turing test has dumb anyways.

=> More informations about this toot | More toots from inertia@pone.social

Written by uis on 2024-12-27 at 00:34

@inertia LMAO. I don't think I will see AGI in 10 years. Probably I will not see it in my life.

=> More informations about this toot | More toots from uis@pone.social

Written by Cesium on 2024-12-27 at 00:35

@uis That's as dumb a claim as saying "AGI in 2 months". You just don't know. No one knows.

The current paradigm will never produce AGI, but we could find something new at any point. Humans proof AGI is possible. Nothing that's possible is unachievable.

=> More informations about this toot | More toots from inertia@pone.social

Written by uis on 2024-12-27 at 00:49

@inertia Eeh. MASes have potential to be foundation of AGI. Yes, heavy abuse of statistics doesn't sounds like something that will produce AGI, but I think of other reasons:

1. Promise of AGI sells. Result does not. Corporate greed will not allow real AGI.

2. Goalposts will be moved. It happened with "intelligent machines"(computers), it happened with search engines, it happened with knowledge bases. It will probably happen with ML.

3. God syndrome. Corporate mental disorders will not allow AGI.

=> More informations about this toot | More toots from uis@pone.social

Written by uis on 2024-12-27 at 00:58

@inertia of course thread was started because of academic research, so we are not completely doomed to corporate hell, but even saying "we will have AGI in 10 years fr fr" is a bit unrealistic. Good specialized AI? Belivable, achievable. AGI? Probably no*. Will be what now exists only in papers be norm? Maybe, just like current AI models and methods were just papers 10 years ago.

• MAS might as underdeliver as LLM

Also "in my life" in my case might be not as long. You know, voenkom never sleeps.

=> More informations about this toot | More toots from uis@pone.social

Written by Cesium on 2024-12-27 at 01:02

@uis >academic research

Is also stuck in the current paradigm. Research has 2 phases: Exploration & exploitation. We are deep in exploitation. Only after the current paradigm fails hard will we get back to exploration and make any meaningful progress.

Everything we do right now is hyperspecialized on a fundamentally wrong approach.

99% unusable garbage that won't be relevant once the paradigm shifts.

=> More informations about this toot | More toots from inertia@pone.social

Written by Pawlicker 🐾😹 on 2024-12-27 at 01:04

@inertia @uis at least the last time the AI bubble popped some nerds would obsess over lisp machines and the lisp language and act as if it was the holy grail of nerd shit

=> More informations about this toot | More toots from PurpCat@clubcyberia.co

Written by djsumdog on 2024-12-27 at 01:07

I've had The Little Schemer on my bookshelf for over a decade. It's traveled through fourteen countries with me and I have yet to get through the 2nd chapter. .. I thought one day I'd learn Scheme or Lisp (I swear I will get to it, damn it!).

Then I did devops for three years a a place where all the devs used Clojure ... because what's not to love about taking all the guarantees from a statically typed language, flushing them down the toilet, running the entire VM in a REPL for production and calling that a good programming language?

=> More informations about this toot | More toots from djsumdog@djsumdog.com

Written by djsumdog on 2024-12-27 at 01:04

AGI? I doubt we'll ever see non-garbage self driving cars within the century.

=> More informations about this toot | More toots from djsumdog@djsumdog.com

Written by スカイラー :fillyDab: on 2024-12-27 at 01:07

@djsumdog @inertia @uis self-crashing cars, on the other hand, are almost guaranteed

=> More informations about this toot | More toots from skylar@misskey.yandere.love

Written by djsumdog on 2024-12-27 at 01:10

AI will never be able to out crash the masters

=> View attached media | View attached media | View attached media

=> More informations about this toot | More toots from djsumdog@djsumdog.com