Overall, what are the differences you've seen between GPT 3.5 and GPT 4 so far? Is it really more reliable?

[shukla7789]

Laurent J. – First of all, there are three ways to use GPT4: have ChatGPT Plus, use Bing Chat, or be on the Waitlist. API access was only unlocked yesterday (March 16) around 6 p.m. I've been testing it since the launch via ChatGPT, and there's been a lot of progress in the content produced.

There are no noticeable differences in some prompts, but the content seems a little more qualitative to me . Some expressions, certain turns of phrase allowed us to detect that the content was generated by an AI, here it is less the case.

Louis B. – The main difference is a clear increase in the quality of the language and the effectiveness of the answers provided . The iran mobile database intention is still just as well understood, but I observe a clear improvement in the quality of the answer: more educational, much more relevant with better-chosen examples, popularizations or elaborations.

GPT3 tended to repeat prepositions a lot more or in bulleted lists (it titled each item in the list and repeated the title right after in an explanatory sentence). Similarly, the old model too often showed in its phrasing that it was better at English. I tended to translate generations. This is less and less the case.

In my numerous uses over the past week via ChatGPT, I have much less rewriting work to do on the proposed results. It becomes more difficult to distinguish its responses from those of a human.

On the downside, the new version is slower, and builds stop more often mid-response. This makes it less usable for automation, but it's only a matter of time before performance improves.

Yannick B. – My comparative study of GPT 3.5 and GPT 4 focused mainly on SEO needs and uses. The overall assessment is that there is still a real gain in quality in terms of responses from GPT-4, but the main thing seems to lie in the ability to understand the instructions and therefore then apply them. Thus, for simple prompts, we will not necessarily see many differences. The real gain will be with elaborate prompts. So no more playing fast and loose with GPT, we will have to work on our prompts .