they probably done this so you cant make your own gpt
Summary created by AI.
The discussion revolves around the issue of ChatGPT not finishing its code or sentence. The initial solution given is to use ‘continue’ or ‘continue in a codebox’ to prompt the model to complete its response. Some users were successful with these commands, while others found that the model still generated unrelated or duplicate code. The problem was attributed not to ChatGPT itself, but to the constraints of the chat interface. Users pointed out that the character limit imposed by the website makes it challenging for ChatGPT to pick up from precisely where the code was truncated, causing disruptions in the code’s continuity. Suggestions to tackle this, like exporting code to .txt or breaking the answer up without altering the data, were made.
Victor.hinojosa identified that this shortcoming is particularly problematic for programmers because minor alterations can break the functionality of a code. Some users speculated that the issue is a deliberate limitation to encourage users to opt for a premium version of the platform. RonaldGRuckus explained that it’s not a marketing tactic but the result of token lengths. He mentioned that GPT-4 offers longer token lengths in its paid version, likening this to paying more for larger ice cream. Further inquiries were made on whether a paid version would solve the problem entirely. Finally, acityhussler found a reliable workaround for the issue in the ChatGPT Plus version when dealing with sizeable codeblocks, without specifying what the solution was. ref
Summarized with AI on Aug 01 2023
AI used: gpt-4-32k
I pay for it and have the same issue sometimes worse. it forgets what i asked him to do 5mins prior
Summary created by AI.
Multiple users in the thread titled “ChatGPT sometimes not finish its code or sentence, how do I solve this?” discuss issues with ChatGPT not completing its code or sentence output.
In post 1, shrusti.eminent initiates the discussion by raising concern about this issue. logankilpatrick advises in post 3 to request the AI continue its sentence and encourage feedback. ic202 simplifies it further in post 4, suggesting the command “continue.”
Rowbot in post 5 shares a method that has worked for him: using “Continue in a codebox.” He also suggests a particular function can be continued by saying “continue in a codebox from def XXX.”
Further down, pealco821103 in post 7 and harwotakjan98 in post 8 express their frustration with the AI, as it tends to generate unrelated code despite the context, and even after requesting for continuation it starts over from scratch, respectively.
Victor.hinojosa, in posts 10 and 11, blames the chat interface’s limitations for these issues, not ChatGPT’s programming. He points out the difficulty of breaking up the AI’s response into chunks, given its tendency to alter or restart the code, and requests a solution.
Castel in post 12 and RonaldGRuckus in post 13 debate the motivations behind the limitation. While castel suspects it’s a strategy to incentivize paid subscription, RonaldGRucKus explains it’s related to “token length.” RonaldGRuckus further reveals that the upcoming GPT-4 has longer token length options.
Iceagod in post 19 and alexbelotsky in post 30, as paying customers, express doubts and share their experiences of cutoffs despite being paid users.
Lastly, acityhussler in post 34 and simonhand80 in post 43 share their frustration with the inconsistent nature of “continuation” prompts, concluding the usefulness they once found in the bot has now diminished.
Summarized with AI on Nov 24 2023
AI used: gpt-4-32k