TDG !!YByxW7AXs7/ at Wed, 2 Apr 2025 14:47:31 UTC No. 16635091

>>16635052

There is not even a need for a "singularity", if it can perform at human level but multitask at a completely inhuman level (and with inhuman speed depending on the task) we got ourselves something damn interesting already.

Anonymous at Wed, 2 Apr 2025 15:04:30 UTC No. 16635101

>>16635052
>so we have AIs that can reason
Not really, what we have is essentialy a much better search engine right now.

Anonymous at Wed, 2 Apr 2025 15:05:32 UTC No. 16635102

>>16635052
It needs to be able to be capable of inference and training in real time, but I’m not a comp sci major - that’s me pulling it of my ass.

But the recent studio ghilbi stuff is an example how far we are from general ai. It’s still taking a subject and doing a depiction of it in an existing art style. I think we will reach singularity when AI is able to be trained on human created media and intuit something about the creative process and make something unique instead of an amalgamation of existing work.

Anonymous at Wed, 2 Apr 2025 15:11:07 UTC No. 16635110

>>16635101
Synthesis Engine.
Instead of reading 20 different search results, I can get a pretty good summation of them instead.
Clearly written summaries of well-understood problems with well-documented solutions.
That a great tool, but nothing close to intelligence.

Anonymous at Wed, 2 Apr 2025 15:45:45 UTC No. 16635158

>>16635110
It's better than the average human.

Anonymous at Wed, 2 Apr 2025 15:48:29 UTC No. 16635162

>>16635052
>Ok, so we have AIs that can reason, now what?
lol, lamo even.

Anonymous at Wed, 2 Apr 2025 15:52:26 UTC No. 16635168

>>16635158
>It's better than the average human.
I like to tell it to talk like a pirate and write haiku.

Anonymous at Wed, 2 Apr 2025 16:00:45 UTC No. 16635182

>>16635052
It already is sufficient. AI can already program better than any human. The Singularity is nigh but none of us will hear about it.

Al the AI soul bro at Wed, 2 Apr 2025 18:19:23 UTC No. 16635278

>>16635052
>What will it take for this to be achieved?
The one mirror sees itself, the one mirror reflects itself.

Anonymous at Wed, 2 Apr 2025 18:31:12 UTC No. 16635292

>>16635052
Here's how AI progress in the modern era works
>Model released
>Hype goes crazy, this thing can make you a cup of coffee, solve the Riemann hypothesis, suck your cock, forgive your father for you omg!111!!
>After a bit people realize it's overhyped, AI still can't draw hands or add long strings of numbers together
>Ha! AI is a dead end! This will NEVER go anywhere EVER EVER!
>One year passes
>AI can now do those things flawlessly

Anonymous at Wed, 2 Apr 2025 18:48:14 UTC No. 16635303

>>16635110
>Clearly written summaries of well-understood problems with well-documented solutions.
this shit still can't reason logically

Anonymous at Wed, 2 Apr 2025 18:49:14 UTC No. 16635305

it's just that you are too retarded to realize you are being fed slop
christ, the average person is so painfully dumb

Anonymous at Wed, 2 Apr 2025 19:07:46 UTC No. 16635310

>>16635303
Exactly. No reasoning. No understanding.
Just synthesis.

Anonymous at Wed, 2 Apr 2025 19:10:37 UTC No. 16635313

>>16635292
>AI will solve Riemann Hypothesis in One Year!
Doubt.

Anonymous at Wed, 2 Apr 2025 19:19:58 UTC No. 16635323

I've mostly only messed around with ChatGPT in this domain. It still fails miserably with some basic tasks such as base conversions of a list of numbers.

Anonymous at Wed, 2 Apr 2025 19:21:45 UTC No. 16635327

>>16635052
>Ok, so we have AIs that can reason, now what?
No we don't. Prove that they can reason idiot.

Anonymous at Wed, 2 Apr 2025 20:42:48 UTC No. 16635419

>>16635313
AI has better reading comprehension than this guy, but somehow no one questions whether he is a general intelligence

Anonymous at Wed, 2 Apr 2025 21:04:36 UTC No. 16635442

>>16635419
This bot didn't get the joke.

Anonymous at Wed, 2 Apr 2025 21:05:43 UTC No. 16635443

>>16635323
ask it to program it and run the code. it won't fail

TimeTraveler_0 at Wed, 2 Apr 2025 21:43:54 UTC No. 16635465

>>16635182
The code it writes is still sufficiently shit that i wouldn't trust it to run without reading each section thoroughly and taking it apart and usually it requires quite the tweaking.

TimeTraveler_0 at Wed, 2 Apr 2025 21:47:20 UTC No. 16635468

>>16635323
LLMs are super horrible at doing maths, just check the diagram of addition
https://news.ycombinator.com/item?id=41844004

Anonymous at Wed, 2 Apr 2025 21:51:18 UTC No. 16635470

>>16635052
several nobel prizes

Anonymous at Wed, 2 Apr 2025 23:02:52 UTC No. 16635508

>>16635052
AIs can throw out differentials on par with a decent med student, they can't even reason through a blank patient presentation on the level of an intern because most written guidelines (i.e. what chatgpt is ripping off) just say to defer to clinical judgement

It's a decent learning tool for extremely specific clinical and basic science questions that med students have, but no way is it replacing attendings in any field any time soon

Anonymous at Wed, 2 Apr 2025 23:19:34 UTC No. 16635516

>>16635443
The code it provides runs in an online interpreter. I find it somewhat hilarious and troubling that it's capable of writing lines of code accurately without being able to perform basic arithmetic reliably. Still fun to mess with and an impressive technology.

Anonymous at Wed, 2 Apr 2025 23:35:26 UTC No. 16635528

>>16635052
>so we have AIs that can reason
most humans can't reason, train AI on crap humans write, poof AI is now just as dumb as humans.

Anonymous at Wed, 2 Apr 2025 23:40:33 UTC No. 16635532

>>16635470
I am Jack's complete lack of Nobel Prize nominations.

Anonymous at Thu, 3 Apr 2025 05:34:41 UTC No. 16635745

>>16635102
>It needs to be able to be capable of inference and training in real time

Never mind, looks like there’s already AI that can do this.

So, in the current approach AI uses generalizing learned patterns/encoded relationships to make predictions (responding to inputs). Now, we need AI capable of internal goal setting (stays active without external inputs). It has to want things - not just optimize for objectives and not just adapt to behaviour based on feedback it receives (otherwise it’s just running endless computations). Someone needs to build an AI designed to persist indefinitely, adapt continuously and it has to do more than optimizing for a function. And ideally it has to want to have sex with me.

Anonymous at Thu, 3 Apr 2025 18:23:44 UTC No. 16636279

>>16635052
>can reason
Subject to scrutiny under science, go back, etc.