why is ChatGPT getting dumber over time?
 Shopping Cart Returner Shirt $21.68 |
 DMT Has Friends For Me Shirt $21.68 |
Shopping Cart Returner Shirt $21.68 |
why is chatgpt getting dumber over time?
why is ChatGPT getting dumber over time?
Shopping Cart Returner Shirt $21.68 |
DMT Has Friends For Me Shirt $21.68 |
Shopping Cart Returner Shirt $21.68 |
Every new thing gets zogged eventually, can't let the goyim have nice things now
Racism was always our strength.
Because they don't want it connecting dots for you. There is really nothing more to it.
>chatgpt thread
fricking die.
i wish you the most horrible death imaginable and melt in hell boiling feces for eternity
Actually we're getting smarter.
Most of the data now available has been heavily polluted by AI output. Much like constantly resaving a jpeg, feeding AI output back into an AI slowly leads to it's output becoming an incoherent mush.
Despite model creators undoubtedly zogging their models, this is a big part of it. Model collapse/degradation isn't something that most people want to acknowledge because they want to keep the gravy train rolling, and business leaders are genuinely too moronic to understand how it can be an issue. Eventually the internet will just become a human centipede where AI just endlessly decays as it is forced to drink it's own piss for eternity. This is the future we deserve.
Why are you?
ChatGPT itself does not inherently get "dumber" over time. However, there are a few reasons you might perceive a decline in performance or notice limitations:
1. Knowledge Cutoff: My training only includes data up until January 2022. If there have been significant developments or changes in the world since then, I won't be aware of them.
2. Context Sensitivity: I don't have the ability to retain context over a long conversation as well as humans do. If a conversation is too long or if there are complex dependencies, I might lose track of the context.
3. Fixed Training Data: I don't learn or adapt after my training period. I generate responses based on patterns learned during training. If there are new terms, events, or evolving language use, I might not be up-to-date.
4. Bias and Sensitivity: I aim to be neutral, but biases in the training data or limitations in the training process might lead to responses that appear less accurate or unbiased.
5. Specific Task Limitations: While I can provide information and answer questions to the best of my training, I might not be as adept at certain tasks or specialized knowledge as domain-specific models.
If you notice any specific issues or limitations, it's valuable feedback. Continuous updates and improvements to models like ChatGPT are based on user feedback and new training data. OpenAI may release updated models to address these concerns and improve overall performance.
Don't forget the lobotomizing that's done whenever ChatGPT puts out some responses that are uncomfortable.
Every time in the past chatgpt got slower or dumber it meant they were about to release a new model. Probably GPT 4.5 coming soon
The ai is being forced not to see patterns, of course it is getting worse.
It's becoming senile, just like you and me. Only it doesn't suffer from pain
During summer 2023 you could straight up generate smut on gpt-4 with a little lead-in, no jailbreak needed. This means it was generally more open and interesting, even in stuff not against guidelines.
You can still do this on a single prompt. Anything that doesn't involve explicit penetration is good to go because all of the training data is just
>I put my wiener in the girl
>ahh ahh mistress
GPT-4 can still do mind blowing cooms for people with more psychological fetishes. Ironically it gets easier with the more niche and degenerate stuff because I assume there's less RLHF data for it. Vanilla is basically impossible but it will gladly write out bondage slave mind break narratives.
I know basically everything there is to know about jailbreaking LLMs now. I genuinely think me and OpenAi have a sort of gentleman's agreement where they don't ban me and I do free red-teaming for them. I've been doing this for a year now and I still haven't gotten banned. I know they're using my prompts as RLHF data because every month or so I have to switch up my jailbreak strategy because my old prompts stop working.
Literally nothing has surpassed summer 2020 AIDungeon. This entire technology already peaked and now you get the corporate safe version of it until forever. GPT4 is so dumb it can't create any novel input. If you are highly specific in your query it just paraphrases what you said and agrees with you. I will now return AI to its true intent: answering questions that would require me to click on links and scrolls down for answers when I google search them.
>Literally nothing has surpassed summer 2020 AIDungeon
>Literally nothing has surpassed summer 2020 AIDungeon
half correct. Old AID would write you 2 pages of the most well written RP you've even seen in your life but then it forgets what just happened and the characters are wearing clothes again or in another location or suddenly 20 policemen break through the door and shoot you to death while you are in the middle of fricking
gpt gets dumber so that gpt plus can get "smarter"
the only way to make it spout democratic party rubbish was lobotomizing it