Should OpenAI (REMOVE -or- KEEP) the “CONTINUE GENERATION” button from GPT-o3? (Yes/No)
-2/24/2025*
Why it’s an issue:
The “Continue Generation” often leads to incomplete or forgotten responses. It feels unnecessary , Instead of improving the experience, it adds frustration and forces users to manually continue interactions that should flow seamlessly .
If you answered No (KEEP it), why do you think it’s useful?
Take out the useless “Continue Generating” button from all GPT models? because obviously, we want GPT to finish generating our messages! right?*
*OpenAI * is a multi-billion dollar company that creates the world’s leading LLMs , and ***adding things back from 2023 into a working system is insane… *** imo.
The “Continue Generating” button disrupts the flow of conversations and often causes GPT to not complete or even forget parts of the message. I was having a great conversation, and then, all of a sudden, it started asking me to continue the generation. Since then, I haven’t gotten a single response remotely close to the precision I had before. My experience was stable, and losing GPT-o1-mini was already a blow, but now we’re dealing with this ‘Continue Generation’ button that we haven’t seen since mid-2023. GPT-3 had this feature, and it makes sense to see it again with GPT-o3, but honestly, it needs to be taken out.
- Token Limits : While GPT has a token limit for each individual response, it doesn’t automatically extend the token limit by pressing “Continue Generating.” Instead, it just picks up from where it left off, continuing to generate more text. The token limit is still bound to the model’s architecture, and once it’s hit, it requires this manual intervention to proceed. So, in theory, it doesn’t extend the token limit but merely continues the process from the same point.
- Why Pause? : The pause often isn’t about “thinking” or “processing” for GPT—it’s more of a system design choice. It’s likely related to managing the flow of responses within certain constraints, like processing power, response time, and preventing too long a response in one go. The system needs to stop and allow the user to decide if they want more generated text. But that doesn’t make it a better experience for the user—especially if it’s creating frustration or unnecessary interruptions.
- Longer Responses : As you mentioned, GPT used to generate long responses (like 3,000+ lines), and there was no need for such interruptions. It makes sense that the model should be able to provide large chunks of information in one go. If your response is under the token limit, there’s no reason it should be interrupted. The feature likely doesn’t add any meaningful value when it comes to long, detailed responses—it’s just a bottleneck in the process.
- Resource Management : The pause might be an effort to manage resources, but if it’s causing more inconvenience than benefit, then it’s definitely a flawed feature. If GPT was capable of producing long, complex answers (like 3k lines), it should just generate them without pausing midway—especially when the token limit isn’t the issue.