[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1713992812482113.jpg (79 KB, 1024x742)
79 KB
79 KB JPG
I've been using chatgpt 3.5 free online demo recently. Is this really supposed to replace people?

>very wordy, takes 5 bullet points with a paragraph per bulletpoint to "answer" my question
>blatantly wrong information
>information that doesn't actually answer my question
>lots and lots of fluff

I must be using it wrong, right? Surely this thing isn't what people are hyping up is it? I mean it's kinda cool but it's more like a google search 2.0 than an AI.
>>
Anyone using an LLM to replace jobs is hopefully using the most advanced version. (3.5 is ancient, anon)
>>
>>100182571
what's the most advanced one? I want to try it
>>
It's fairly over-hyped. I'm not even convinced LLMs have improved much. They just lobotomize them slowly so they seem more impressive on the next iteration.

I will say they're a great help when it comes to professional writing, sorting through documentation, and tasks that require large amounts of knowledge. They aren't the spooky scary boogie man that people act like they are. Not for at least another 30 years minimum. If that.
>>
>>100182582
Either ChatGPT 4 Turbo (newest one, April 2024) or Claude Opus are top tier right now
>>
>>100182448
They made it wordy exactly to distract people from the low quality of the content.
Most people are easily impressed by long sentences and don't even pay attention to what it's actually being said.

Politicians often employ the exact same tactic.
>>
File: glassesloopy.jpg (85 KB, 716x955)
85 KB
85 KB JPG
>>
File: 1684614214624977.png (755 KB, 1010x719)
755 KB
755 KB PNG
>>100182582
>>100182652
This. Also you're not using it right. You're not supposed to ask it simple questions. You're supposed to add a bit of context. I figured this out very quickly but then again I'm not a retarded, objectively inferior subhuman like OP who posts ugly shit like pic
>>
>>100182448
>>100182701
>>100182708
i love asiasn women so mcuh fuckkk
>>
>>100182448
>very wordy, takes 5 bullet points with a paragraph per bulletpoint to "answer" my question
You need to tell it not to do that.
>blatantly wrong informarion
On certain topics hallucinations are worse than others

Overall, it sounds like a case of severe user error.
>>
>>100182701
honestely?
id fuck
>>
>>100182708
>>100183078
Why do I need to tell it not to be wordy? I tell it not to be wordy and it just shortens it's wordcount but doesn't answer my question. Almost seems like it's vomitting out a ton of information hoping some of it is correct.
>>
>>100182448
use gpt-4 and use these custom instructions for an actual smart AI companion
>Talk to me like a person, not a list generator. Keep it professional. Keep answers short unless asked to elaborate. Give a complete answer, but be concise. Avoid list format unless absolutely necessary (as in, I explicitly ask you to generate a list).
>>
>>100182701
would.
>>
>>100182701
Fuck, why is it so hard to get women that look like this in the US, I just imagine how wild in bed they must be too, totally crazy
>>
File: IMG_2207.jpg (10 KB, 225x187)
10 KB
10 KB JPG
>>100182448
Everything it says is generic, zero personality, corporate, family-friendly, morally neutral reddit faggot shit. Try talking to it about any topic with weight and it starts giving you suicide hotline numbers. It can do some interesting things, though! These include:
>correcting simple code with roughly 62% accuracy
>pretending to be your anime wife who forgets your name and what you were talking about nine minutes ago
>letting Rajindahar sound like a real person in spam emails
>>
>>100182448
Normalfags just using it for basic stuff won't understand. Only those of us using it for code understand how well it's able to think logically. 3.5 was mind-blowing, one of the most amazing pieces of technology I've ever used. 4 is a nice upgrade.
>>
>>100182448
>chatgpt 3.5
that's ancient anon
>>
File: gpt4_and_pixi_js.webm (3.96 MB, 640x480)
3.96 MB
3.96 MB WEBM
>>100182448
>>
>>100182448
>I must be using it wrong, right?
You know, most people prefer wrong information.
>>
>>100182448
you arent being explicit enough
>>
>>100182746
>>100183170
>>100183300
>>100184888
Please have some self-respect. I want to vomit looking at THAT.
>>
>>100182701
Repulsive
>>
>>100182701
Adorable
>>
>>100184991
cool
>>
>>100182701
the post that tore /g/ apart
>>
>>100185089
just say you're GAY
>>
>>100182701
would
>>
ChatGpt really isn't that good. I know guys who use it for coding but they also suck. If you are competent it's not a huge help. And yes, I'm using gpt4.
>>
File: 1712690552966855.jpg (138 KB, 1080x1350)
138 KB
138 KB JPG
>>100182448
>>100182701
>>100182708
>>
>>100182708
Don't bother, OP probably types in all lowercase letters with no punctuation, no structure, and doesn't frame his request as a story.

>>100183178
>but why do I need to [format my input to get a certain output]?
Because that's how human beings talk to one another and that's what the training data is made of.
All the text that the linear regression algorithm used is human conversational text shit, not 500 terabytes of Table of Contents entries.
Although I bet if you could write like a Table of Contents entry you'd get good results too.

The more "context" (aka prompt text that aligns with the thing you want) you provide, the more the algorithm can zero in on the information you want.
The more vague you are, the more you leave out, the larger the range of mathematically valid answers you get. As you provide more text that relates to your query, you narrow down the scope of probabilistic correct responses.

It's just a big search engine, or like a huge answer key to a test, or a rubric, or whatever.
>>
>>100185508
CUTE!
>>
>>100182448
It's perfect for future humans.
>>
>>100182701
I'm 30+ and I'll take whatever tight asian vaginaI can get
>>
>lust provoking image
>>
>>100186788
Checked
I wouldn't be surprised if one of these days I start fapping into anthropomorphic caricature pornography
>>
>>100182448
>chatgpt 3.5
Cease to reproduce if you really can't even look up what the most up to date gpt version is before forming an opinion
>>
File: queen and king of g.png (200 KB, 255x516)
200 KB
200 KB PNG
>>100182448
>>100182701
>>100182708
>>
>>100183178
>tell it to not be wordy
>its supposed to automatically know what you mean by "not being wordy"
kek retard of the year
>>
>>100185089
>gay or landwhale lover
Confess.
>>
>>100182448
I got the same shit from Llama 3 70B when I was checking that out (I don't use LLMs). Nearly everything was bullet points. Kinda lame.
>>
>>100182448
>buddeng technology is seen as it's maximum potential
Gee willikers anon, I wonder why we aren't riding the hottest 1900 model of cars
>>
>>100182448
You must be retarded. I've pretty much replaced 90% of my googling habits with chatGPT-4. I have a very good intuition to the kinds of questions it can answer and I didn't even need to learn anything.

Just so we can laugh, what question are you asking?
>>
>>100187459
the machine is supposed to speak fucking english telling it to not be wordy should be enough
>>
>>100182448
No. Most of them suck. The Bing one just rips the bullet points out of the first 3 searches. Its only good for lonely autists and coomers.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.