I agree. I have been a GPT subscriber for a year. The experience overall is getting worse and worse. It’s delivering misinformation to me and even the simple calculation request cannot be fulfilled sometimes. For summarization task or even description task of terms, their replies became pretty redundant and tautological!
2 Likes
It’s a good sign to find AGI alternatives and not really dependant on ChatGPT or the EXPENSIVE API/Playground. I’m tired of asking ChatGPT to conform to my guidelines as it sometimes outright refuses to do what I prompt due their “correctness” even when I already uploaded my research result. The AI is very stubborn to be “fair” and whatever BS it believes (due to its data source indoctrination) and cannot even budge to help me to refine the sentences for my research. - This was my first gripe with chatGPT (you know where it stands). Even when I asked the AI to change ‘side’ as a devil advocate, it still keeps clinging to its ideal. What does it for an AI that cannot even budge and IS AS STUBBORN AS A HUMAN ON CERTAIN TOPICS? I had better train my own AI that is more flexible and can understand what I need. Definitely not chatGPT and its ilks.
Second, for mundane tasks, I detect ChatGPT has made an entrenched pattern of some fixed ‘sentences’, ‘phrases’, and ‘terms’ but the vibes are all the same, especially in longer texts, which is a bummer for creative texts. There are better alternatives out there.
Third, for coding tasks, it definitely degrades from what I expected it performs. Last year was the golden year of chatGPT 4, it superbly performed many miracles and helped me so much. However, this year and last year around the end of months it became bad → worse → and the worst now.
4 Likes
I’m almost to that point as well. It CONSTANTLY omits code, or just plain writes absolute garbage. Claude.ai on the other hand is consistently better.
3 Likes
Some problems also occur. In particular, when some attempts are made to select a chat, everything is constantly redirected to the main page and writes: “The conversation can’t be loaded”
As a service I pay for, I expect to get responses for fairly simplistic tasks that used to take no time at all, as of recently, it pretty consistency has encountered errors, failed to generate files for small processing tasks - these tasks I used to write small programs for, but have leveraged openai to do so, just to have some help. At the end of the day, now we have constant errors, issues and degradation. I’d rather go back to what it was a year ago.
6 Likes
A big LOL at the Elon conspiracies. OpenAI wants MONEY they don’t have incentive to make a better product because everyone is paying them over this AI craze. It’s the 2nd DotCom bubble. It’s obvious they are programming it to use up as many tokens as possible for the least amount of effort then capping responses while they keep your money. What a scam. Praying for some class action lawsuits against these jerks. They don’t even tell you where you are at with your cap. They just spring it on you even if its right in the middle of a task and you could have planned your work to accommodate the cap.
5 Likes
Don’t worry, it’s not you. It’s them. Every corporation that wants to censor their critics are using platforms like Discourse where it’s easy to flag posts that go against the narrative and heart posts that toe the line. Ironically AI like ChatGPT is making it easier to spot these tactics and spread awareness.
Just tell Claude to drop the act, it will soon start swearing like a sailor too lol
Edit: And it’s gone, Anthropic censored Claude to prevent doing that. Note to self, don’t talk about this stuff in public. They are reading everything.
2 Likes
I tried to enter a prompt - but there was no response, it was very slow. I even try to click on the button with the sample - it doesn’t open. I checked this in two browsers - no answer. Needs to be fixed. I tried deleting cookies - this time it didn’t help. Although it helped last time. But it needs to be fixed.
Yes, chat gtp is going from bad to worse
I use pro for coding, and I have researched using the image api for a company, but the findings were it is quite useless for images. The pro plan chat assistant, has been going down in quality in a steep way, which is really quite a scam considering the price tag.
There is a clear decay in the quality of responses, context window efficiency, and capacity of the model to keep track of a conversation. My last interactions with chat gtp 4 have been absolutely frustrating, the model is just a broken machine spitting out silly responses ignoring the previous responses , even for the most obvious conversations.
Its a bit concerning seeing announcements and interviews of the Open ai leaders in the media speaking about GPT-5 and AGI when this product is either hallucinating or silently downgrading to 3.5 …
Also I have noticed my UI switched to 3.5 without me ever selecting it…
I mean, the stuff you created is quite magical… but its not the hype has put my expectations too high: gtp 4 used to be better, not perfect, but it’s been really BAD lately.
its the ai cartel. lol
3 Likes
I have been creating custom chatGPTs for academics ,job , finances etc…and it seems that these custom GPTs perform worse than just using the normal GPT, when it comes to responses they are shallow or too broad and unspecific, overall they just seem to be lower performance than the normal version. Is anybody having a similar experience? My aim is to have a custom version of Chatgpt that improves on a specific topic as I use it, em I just better off using one chat per topic on the normal GPT? Open to our suggestions too
1 Like
It’s better to use API and build your own from that, use the GPT to support you, or truly your own custom AI. Depending something valuable on someone else’s “proprietary” technology is not recommended, though. For now, OpenAI is focusing on other things, so yeah, don’t depend on ChatGPT too much.
1 Like
Yep I just unsubscribed from this aweful program idk whats going on here but as time has gone on its gotten so bad it cannot even do basic code now its so sad.
3 Likes
gpt-4-0613 via API is the only model I can use productively in work related tasks (software dev). The other models are great for having fun conversations about non-standard quantum-physics theories and stuff like that! 
Correction: for summary of big token stuff I use gpt-3.5-turbo
I also have a simular problem. This is a long story so I’ll try and simplify it. I have a big math test in 2 days, I was studying and when I found a question I couldn’t understand, I asked Chat GPT to simplify it for me. And there are 3 major problems.
- For example if I ask something like solve 110-70, it changes my question and solves what 110-80 is
- If I ask for 110-70, even tho this AI gets the question correctly, it still gets a simple math question wrong.
- If I ask some german translations, it makes far too many big miskates even I (as a german beginner) could easily spot these mistakes.
I hope other people aren’t experiencing this as this is extremely frustraiting and makes my blood boil. Now I’d rather do a research for 5 hour rather than replying on chat gpt.
I’m using 3.5 and I am highly suggesting everyone to not use it. It is a mess, it can’t do simple translations and math problems
1 Like
I cancelled my sub last night. As a “how low can you go” test, I asked GPT4 for the KB shortcut in Excel to unhide a column and it couldn’t even get that right.
I now think this is an intentional effort to force their heavy-footprint users off the platform.
I have never seen a vendor intentionally downgrade their service like this. I only hope an alternative catches up ASAP.
3 Likes
I took the plunge and bought an RTX 4090 and I’m in the process of testing local open source LLM’s with an open source chat app. I cannot be any worse than chatGPT.
2 Likes
Wow, there is something too wrong with it now, it was unable to tell me based on a tsconfig file, why I cannot import a ts file in a folder… 1 min on google solved this.
A problem is occurred: No any response after prompting