A representative 300 sample would give a more accurate result than a biased 2.4k sample. Bigger number doesn't mean better results.
That said, I'm not sure how to get representation from certain subgroups of the population, like the "never engages with polls" or "lies specifically to fuck with your data" subgroups.