Those 'amazing' GPT-3 videos only gloss over the good side of what GPT-3 can do given an input. They'll never tell you the number of times they cherry picked and edited the output or the mistakes it makes nor can GPT-3 itself reason of why it has generated that output or chosen those words in the sentences it generates.
If a black box neural network like GPT-3 still can't explain itself, then it cannot possibly safely help "humanity fight COVID-19" or any disease if you show the generated results to a medical professional.
So beware the hype brigade.