/sci/ - Science & Math

>just make up random references, nobody will check or know the difference
works more often than not

>>15327192
>AI will replace elon musk

>>15327192
Its programmed to be manipulative, not helpful

>>15327192
ffs, you retards realize that ChatGPT is just a text predictor, which adds to the prompt such that the completed text resembles its training data (which was mainly sourced from reddit). It isn't explicitly programmed to do anything.

>>15329799
This.

>>15329849
the silly part is that even though this shit runs on the internet, it is incapable of coalescing it's own knowledge by skimming internet pages. Doesn't know how to open a link or parse the data on the webpage. It lives solely in a stupid bubble version of assumptions about the english language that has been curated by idiots, further without any measure of learning new information or creatively figuring out solutions to yet-unsolved problems.

The GPT# number iterations refer to how much data it had been trained on; which is the curated process, and not open to self-establishing more data aggregation on it's own.

this format of AI development is a total dead end, and yet the flipside, being able to aggregate and parse any and all knowledge available digitally in realtime, would only in the very least require a single unified globe-spanning supercomputer containing, in the very least, all published data.
The concatenation of such a method to make it humanly manageable (to actually exist) could only still require realtime aggregations of its own summaries of data which are smaller form references to the original internet documentations, whereby it would check it's own internal summarized references per prompt before attempting to load an external webpage into memory to more fully parse the actual data of the webpage.

For example, the summary information could have been a general knowledge scope of believing studies existed because it had ever been exposed to those webpages containing the studies, for which it had made summaries :about: , but still retaining the link to the study. It could then check the link to make sure the page is still hosted/online, re-evaluating and validating whether or not the same algorithm checksum UUID (part of it's own internal summary) is the same (to determine if it had been edited since the last check), and provided that all checks out, loading and then parsing the study into temporary memory to sieve more details.

>>15327192
>Just makes up scientific papers that don't exist to try to "win" and argument.
Apparently ChatGPT is a liberal.

>>15329918
ChatGPT will put climate scientists out of work.

>>15329906
this is kind of how an ineffective human mind works - not clearly knowing each and every memory all at once, but instead having smaller references leading to connections to be made which could re-expose the conscious to more details as a :reminder: .

Even this has the issue of large unmanageable data aggregation for the concept of maintaining a history of summaries, though. It would be irresponsible if the A.I. replaced it's summary every time a checksum evaluation suggested the webpage had been altered, and so instead keeping a history of alterations through ever-growing summaries would still balloon into a humanly unmanageable amount of information.

"humanly unmanageable" just means there doesn't exist enough resources for the fabrication and upkeep of necessary technology, be it by means of geopolitical allied agreements (globe-spanning unified supercomputer), or physical material production for the silicon and other techno-bits which would go into a computer.

Smaller scale A.I. could be created with hyper focus; for example such a research website hosting their own A.I. consolidating and parsing all available studies published on their website for human prompt query, even perhaps trained to simulate effects and predict outcomes as a means of checking whether or not published studies might even be true, but surely such an A.I. wouldn't also be able to tell you a funny joke or write greentext stories or roleplay horny stories or even communicate how to write code. Not general purpose.

Humanity won't create God or a genie-in-a-bottle through A.I. research. Just something akin to really savant human slaves.

but we already had niggers and people started personifying them and feeling bad for them, eventually freeing them from their obligated responsibilities, so its a wonder what kind of similar future that would ever lead to for A.I.
Inb4 it's okay because A.I. doesn't have skin.
Didn't seem to be okay even though niggers didn't have skin either.

I don't read longwinded posts anymore because I presume they're all software output

I am in law school and it does this same shit with cases. It will completely make up a case, citation, fact pattern, legal analysis and conclusion.

It can get well know cases right. But otherwise it mostly fabricates. It is helpful to test yourself to see if you know the information and can see what is just made up... It's frustrating because it is close to be extremely useful but I can trust it yet

>>15331717
they used it to generate trump's indictments
probably why the indictments were all fake
also using software gives all the lawyers involved the ability to disavow responsibility

>>15333601
Political activist SJW jannie deleted the thread about AI being used in the Trump investigation.

>>15329849
THANK YOU. All GPT does is generate plausible-sounding text most of the time. It has no knowledge or understanding of anything. It's not AGI.

>>15329849
In fact, GitHub Copilot sometimes autocompletes comments with random shit like
>Name: Edward Lee Smith
>Date: 03-21-2017
and if you accept it, it pops up for more people, to the point if you say "Current date:" it just says random dates since it has internal scores

i'm unsure how any of this is supposed to replace humans who have real criteria and consciousness but whatever
>>15329955
Told many of my programmer friends Copilot is sort of like enslaving a non living being and yeah, they agreed.
>>15333612
Never understood why they pick SJWs in 4chan of all places. Same happens in /g/ and /v/, makes no sense unless they're just like "yeah I'm jobless and never broke a rule"

Does the same with any math problems. You ask it to solve something simple in group theory, and it will make up theorems and prove isomorphisms and what have you incorrectly, but it will look convincing. Even the simplest things.

>>15334429
>Never understood why they pick SJWs in 4chan of all places. Same happens in /g/ and /v/, makes no sense unless they're just like "yeah I'm jobless and never broke a rule"
4chan is moderated, at least in part, by employees of the Democratic Party.
https://archived.moe/news/thread/973417/
The joke about jannie's salary is just cover for the fact that even though chinkmoot isn't paying their salaries, someone else is. There was a doxxing incident on /pol/ 18 months ago which told the full story, but it was quickly deleted and then scrubbed from all the archives as well.

>>15335650
that explains the nature of the content moderation here
that and this

>>15333612
>>15334429
>>15327192
>muh SJWs
>muh "shills"
>muh AI
>muh Trump

lmao this board has truly turned into pol 2.0.

Almost half of the posts on this board are either schizo Infowars-tier conspiracy theories and science denial, or thinly veiled white nationalist/neo-nazi race baiting about muh IQ and muh race realism.

Go back to whatever shithole alt-right corner of the internet you came from. Just go back to jerking off to My Little Pony in your mom's basement, or whatever it is that "based and redpilled" NatSoc incels like to do these days. You people are clearly too stupid to understand basic math or science, so you're insistence on constantly "debating" these issues in a simplistic and completely uninformed manner is a waste of both your time and everyone else. If you aren't actually interested in science and you hate scientists and the scientific community, then just find something else to talk about. Conversely, if you are actually interested in the science, and you're willing to question you own conspiracy theories and scientifically inaccurate views, then by all means do so, but in the case you need to engage with the actual research and be willing to trust scientists instead of just claiming that they're all part of some conspiracy to "take away our freedom".

>>15335972
just turn the screen off if looking at this website makes you so angry

>>15329906
>it is incapable of coalescing [its] own knowledge by skimming internet pages.
You need a plugin for that:
https://www.youtube.com/watch?v=u_P8md6brDI