Any foreign language speakers who have noticed a SIGNIFICANT DROP in the quality of the output of this thing? It can't even generate a simple and coherent text for me in Swedish anymore
Can somebody try it and see?
Ape Out Shirt $21.68 |
DMT Has Friends For Me Shirt $21.68 |
Ape Out Shirt $21.68 |
Look at this very simple prompt:
>Write a text about flying airplanes
Immediately, on the first sentence, there are 5 anglicisms and idioms that aren't used in Swedish. This never used to be the case. GPT has been great for MONTHS since it first came out but now? It can't fricking write a simple paragraph without sperging like a fricking moron
Is this the case for anyone else?
>It can't fricking write a simple paragraph without sperging like a fricking moron
maybe because most swedes can't write a single paragraph without sär skrivning and anglicisms
It honestly wasnt this bad before. I fight with gpt and express frustration with EVERY GEN now, it wasnt like this before. What the frick happened and why am I the only one complaining about this on the internet right now?
google "openai trained by africans" and have a read about who is training the model weights.
The Kenyans are the ones who trained their moderation AI to classify text. They didn't train GPT to generate text. Learn to read moron
It is a valid statement that their input changes what is considered "unsafe". When this output is implemented, this changes the weights in the model. Learn to think, moron.
>why am I the only one complaining about this on the internet right now
Because everyone finished complaining months ago, and stopped using it.
Really? And did they move to something else or what?
claude.ai is 3.5ish tier on most things, and it doesn't complain all the time or use 50% of the answer on 'safety paragraphs'. Local models are viable with a good system and really quite acceptable if you accept their limitations. There are the other big named ones, but I've had limited success with them - perhaps I expect too much.
I tried Bard with no success. Local models are okay but don't really work for foreign languages, at least from testings i did before
globohomosexual nerfing technology because it became too useful for normalcattle, it isn't the first time it happens just look how accurate google search used to be back in 2014
You're kinda late. You should've walked away at the first signs of /misc/tardness.
Have you tried buying Pro version?
Yes that is what I'm using and I've tried all the available models, even the ones in between such as the Turbo models and others
nobody cares about your shitty language, technology is english, nobody will translate into 200 dialects because you skipped english classes to suck migrant dick in highschool Sven
Can we blame Russian troll farms for that multilang nerf?
Somehow there is less Russian propaganda lately, and their trolls used to be unbearable.
>Somehow there is less Russian propaganda recently
meat grinder go brrrrr
Russia is killing and getting killed by former Russians and somehow have been psyopped into thinking this is a win
nafo homosexual, can't wait to see you humiliated when you're force to admit the truth that ukraine is getting their shit pushed in
are the russian trolls in the room with us now?
Yes
I don't use it in my native language because it was always much better in english, but I did notice that when you ask it to regenerate an answer it will shit out almost exactly the same output with maybe one or two different words, it wasn't like that before, it used to vary a lot between regenerations.
GPT-4 is still fine, from what I can tell. GPT-3.5 has been ruined. It is absolutely not as capable as it used to be. No question about it.
Hell it gives me horrible results even in English. Additionally, it fully stopped giving good snippets of code, now it's just stuff that does not work at all.
>new school year
>why gpt won't do my homework anymore
top kek
Took me 7 hours to finally finish the 3500 word text USING ChatGPT.
7 hours. Had I known the subject, I could've written it myself, from scratch, in 35 minutes, but now it took me 7 hours to deal with anglicisms and foreign idioms in order to correct the text. I honestly can't believe it, but I couldn't just end it when I got halfway. This garbage is no longer usable in foreign languages, no clue how it is in English. Actually sad. It's insane that it was only usable for 1 year. Actually fricked up. Wonder how businesses will do when they base everything on OpenAI and then they frick their training up.
Businesses use the raw API moron, not ChatGPT
the API models don't change unannounced and you can use old versions of the model
GPT-4 and Claude2 have huge cuck filters on them thanks to BOT chads creampieing them thousands of times per day for the past few months.
PLAP PLAP PLAP PLAP PLAP
GET PREGNANT GET PREGNANT
GPT-4 is trash. you're paying for something worse than the free version that also has a message cap Lmao
yeah
it also can't write proper code anymore. I had it write some autohotkey scripts and edit some userchrome.css lines and it fricked up. it worked better when it came out.
I just talk to the chat bot every day
Italian works fine.
Got a drop in quality months ago, now It's ok.
They made the free version shittier so you'll pay for GPT 4