However, how does inference work? If you look at it, I have written down the thought processes for 18 seconds. But how did these thoughts come about? What prompts led to them? I'm curious about what kind of thoughts were involved. OpenAI cannot disclose this information; their unique technology does not reveal the inference process. If you try to hack it with continuous questions to break through GPT, they will make it unusable, and they are very strong in blocking this. The reason is that this is OpenAI's unique technology, and if it gets breached, many countries, starting with China, would create similar models. So, they have stated that they won't disclose the inference process. However, I wonder if this is really a good strategy. In any case, OpenAI will not share our technologies.

It seems clear that OpenAI is quite expensive. I showed you the graph; it’s incredibly pricey. We are not developers; we just want to use the ChatGPT Plus subscription. What does this have to do with us? The fact is that the limits are very tight. With the free preview, you can only send 30 messages per week for questions and answers, and with the mini version, you can send only 50 messages per week. Of course, it's tough for developers to use it as well, but even paid subscribers of ChatGPT are currently facing very tight limits. In the future, as various optimization processes are carried out, the number of messages per week will likely increase significantly. However, for now, it is very expensive, and the limits are quite restrictive. There is also the issue of slow speed. Complex reasoning problems can be slow.

However, if it takes a long time for applicants to infer and think about what to eat at a Chinese restaurant today, then there is a problem. The inference process for all questions has to be forced, which results in a very slow final response time and low scalability. I think this might be due to it being a preview version. Without scalability features, it's challenging to try various things. The scalability features I refer to include web searching, web document analysis, YouTube analysis, web file analysis, data analysis using Python, image generation, and creating customized GPT models. These limitations exist in the OpenAI O1 model.

일주일 고작 30개 질문 한도, 오픈AI o1의 분명한 한계는?!

다양한 인공지능 정보를 다루고 있습니다.