I got the question what the difference between the OpenAI Playground “in chat mode” and chat.openai.com is. So for everybody’s benefit, my answer reproduced below. If you’re not familiar at all, see this screenshot for what I am talking about:

Note that this pay-per-use, so a credit card is required (but a free tier seems to exist for new accounts). Bonus: not geo-blocked in Italy, to the best of my knowledge. Anyway, on to my original answer (translated from German to English by GPT-4 😀): The difference is at least the System instruction, which is not visible and not directly influenceable in chat.openai (in my terminology “ChatGPT”). I attribute, for example, the fact that ChatGPT seems “moody” to me: one must first soften the basis of these specifications - or something like that. Also, ChatGPT has these nice formatting features:

… which, however, are cut off for too long answers and, when prompted to continue, break:

… and then it just gets annoying. In the future, there are supposed to be these plugins, through which not only code will be displayed in such a prepared way, but also, for example, WolframAlpha will be directly embedded:

But that is still behind a waitlist. So the Playground is more spartan, but as of now, it is actually the better alternative for professional users - especially if you have GPT-4 unlocked there.

Also with ChatGPT Plus:

(said “default” being GPT-3.5 🙇‍♂️)