The default GPT-3 Playground doesn't have web access, no. But that's a mere contingent fact. You can certainly add in retrieval capacities to large language models (that was a very hot trend last year), and even GPT-3 can browse the web - you may have seen a demo or 2 on Twitter, but OA did it much more seriously recently with https://openai.com/blog/improving-factual-accuracy/ Letting GPT-3 browse a corpus of web pages, picking what links to follow, summarizing, and using them to generate an answer.
(Just one of the many capabilities GPT-3 has been shown to be capable of, which OP will never tell you about, because he's too busy using it in the dumbest way possible and peddling misinformation about eg. what sampling is for - no, Dr Professor Smith, temperature sampling is not about 'avoiding repetition', which is good, because it doesn't avoid repetition anyway...)
(Just one of the many capabilities GPT-3 has been shown to be capable of, which OP will never tell you about, because he's too busy using it in the dumbest way possible and peddling misinformation about eg. what sampling is for - no, Dr Professor Smith, temperature sampling is not about 'avoiding repetition', which is good, because it doesn't avoid repetition anyway...)