A poll is a model, one that samples a dataset of respondents then makes some big assumptions about how representative that dataset is. Such models take time to develop, and are frequently dependent upon previous elections to make estimates of trends. Assuming that some pollsters are not deliberately putting their thumbs on the scales (and I'm dubious that they aren't), the reality today is that this election is VERY unprecedented This means that you have fewer datapoints, greater uncertainty in the validity of those datapoints, and a much bigger margin of error.
One thing that is worth noting - for all this, while the exact values of the polls vary from poll to poll (meaning different predictive models), the variance of any given poll is surprisingly close to zero - and has been for several weeks now. In 2016, the variance was all over the place - people had not made up heir minds, and the margin of error (the MoE) was huge, nearly 4.5 points. Today, the variance is maybe 0.5 points, and Kamala has been consistently ahead - not by a lot, but by enough. This means that big gaffes (like the MSG "joke are likely going to have a very small impact either way.
Final point - a significant percentage of the electorate has already voted, and in many cases the totals are setting records. Are all these Democrats or Republicans voting for Harris? No, but it does mean that a lot of the election has already been baked into the totals, which means that variations near the end are unlikely to have the same impact as they did in years past.