Absolutely, it becomes completely unusable. Funny how it can’t even output more than 150 lines of code, even with 32k tokens. I will not be renewing my subscription.
Exactly, it’s not useful at all right now - I can’t even get GPT4-32k-0314 to give me 100 lines of code, it’s totally ridiculous. Huggingchat’s Mixtral models are much more useful and even free, unlike GPT4 they actually follow my instructions. GPT3.5-0613-16k is the best model (that actually follows my instructions more ore less) openai has right now - which is again ridiculous and shows how much GPT4 has deteriorated. When they introduced the GPT4 Advanced Data Anaylsis mode, IMO GPT4 was at it’s peak. I would like to work with Claude2, but it is not available in Germany. Gemini 1.0 Pro is already much better than GPT4, I am on a waiting list for Gemini 1.5. Honestly, it looks like OpenAI is losing the race, I will not renew my Plus subscription.
Would you be willing to share the rest of the conversation?
22 hrs ago Mistral LARGE: The AI With CRAZY Reasoning Powers??
btw i’m in waiting list. I am looking forward to test OpenAI competitor
for creative writing that’s beyond kids books ages 1 to 4 chatgpt is dead.
Just today I noticed a slight improvement in gpt-4-32k-0314. It followed my instructions and didn’t abbreviate ~200 lines of code. Works well today, I wonder where this sudden development comes from 
Doesn’t exist as I regularly delete my past chats. However I can tell you that it didn’t end well, it had the worst iq drop I’ve ever experienced, total disastar. I ragequit and grabbed 2 months free of Gemini Pro 1.0 and it was a difference like night and day. But today I had a much different (positive) experience with gpt4-32k-0314 and I don’t think this is a coincidence after yesterday’s big news.
If you get some really funky behavior, it would be nice if you could share it so we can either point and laugh at it, or figure out what went wrong 
0314 is probably the best model out there, hands down. It will be a sad day when they retire it. 
Friends, It is getting worse by the day.
-
The instructions you give are not read fully and if ignored for a part
-
The history is so short the model can remember that you can not post any more 150 lines because everything before is gone from memory.
-
I am surprised by the day how it goes worse and worse.
4 90% of the time you are repeating, deleting or need to start from scratch all over again
-
We don’t hear anything at all what is going on. I have a paid subscription, but I would like to pay triple for a working model.
-
Do not think the model from Google is better, it is simply 10 times worse. And completely useless.
-
Who know the exact person responsible for this matter? I would like to communicate, but I doubt it is possible.
-
It seems the result of bigger companies who pay much more and get no restrictions. All the GPU power is going there, I guess. The BING will also absorb a big part. Do not think that canceling your subscription has any effect at all. There are millions of ChatGPT subscribers, and how many complain here? 250 posts, which is nothing compared to the actual users.
If OpenAI was interested in our opinion or feedback, we would have had an answer a long time ago. It seems all wasted time and energy to write about this subject. Kind of talking to a wall or a tree. They also give any response!
Thanks to the world leaders who will prefer not give this kind of tools to citizens, you see this happening. In the end, it is nice to write a letter etc. Coding is a disaster, day by day worse.
Sad to see this happening while it was so good at the beginning!
Agree, it is so bad that it is a pain to use it. Actively looking for other solutions
Same here. Today I suddenly realized I’m spending 2x more time trying to solve a problem or discover a topic when using ChatGPT-4 as a sparring partner than doing it alone.
Looked back on the chat history for the last 12 months and realized the last couple of quarters the conversation started consisting of me mostly telling GPT4 what it missed, asked it to redo things, pointing out where it was lazy and generally having a verbal fight with it. At launch we had a great “relationship” but since about summer or autumn the model crossed a quality line where it became more expensive to interact with even if it had been a free service. The intervals between the chats also increased significantly over time which when looking back at the quality degradation now seems self explanatory.
The last quarter it has also forgotten it has a code interpretor or bing search integration and has started reverting to “Sorry, I don’t have access to the latest info”, or “sorry, I can’t run code”. This means wasting numerous prompts to convince it “yes, you do have these utilities”.
The straw that broke the camel’s back was a 40 minute session today spent with 20 prompts back and forth trying to get the model to translate one of my old ruby scripts into python. It created a main method and then added a comment in there “# implement logic same as Ruby”. It’s like the Owl meme; “Draw two circles, then draw a f*ing owl”. I had plenty of such tasks in my chat history from about a year ago where it just did the job, and did it well, without any admonition required. Fast forward 12 months and after 20 prompts it still fails to make progress, since it just alternates between forgetting the task, the context and/or the objective. The rewrite took me 20 minutes by hand.
Paying money to solve tasks 2x slower at best and requiring my full attention compared to paying nothing is an awful value proposition, so I’m out as well.
PS. For those about to cancel. When you do you get to fill in a questionaire about why you’re cancelling. It’s a great way to provide feedback. Unfortunately that form is likely being distilled by GPT4, which due to the lobotimization of GPT4 means there’s scant chance that any feedback will actually bubble up to the product team.
Hi Jerry
Interesting. How can I activate GPT4-32k-0314 in the CatGPT 4 builder?
Thank you in advance!
In ChatGPT, they give you the AI they give you.
On the API, you can choose the AI model, such as from March 2023 (although OpenAI has gone back and hit some of them with changes) - and pay for it. GPT-4.32k you don’t want to chat with without overarching management of your own past chat turns and input context. It can bill over $1 a question.
Just for your information…
It is not available to activate it in ChatGPT, you need to use a client with an API key, also you need to be at least Tier2 to use the 32k model. I think you can also test API models here: https://platform.openai.com/playground
Your reply in this topic, showing “This post was flagged by the community and is temporarily hidden.” was because community members pressed the flag button on your f* word post.
It is far more useful to demonstrate the AI doing nothing that you instructed it to do.
Like the chat I ended with “The file names were not sent as part of the user text like before, and only if they exist. The whole exchange between you and I has been failure since the point of wanting these thresholds added, where I started with working code and you haven’t produced working code since.”
Ah I see. Thank you for your reply Jerry!
I tried it, but basically it seems to be ChatGPT 3.5. For me, it is not better, but it depends probably on what you are trying to achieve.
Maybe for Python it is a good choice.
I’ve cancelled my subscription. ChatGPT refuses to help me analyze MIT-licensed source code. Absolutely no copyright issues, but OpenAI has decided to nerf this tool into oblivion.
They flagged my last post because I linked alternatives to ChatGPT. Google for them. With OpenAI actively making their tools worse, it’s easy to find better LLM tools from competitors.