Yes, there's a blip at the end of the chart.

But the headline is clearly that deaths have fallen ~70% even as total (reported) cases has continued to grow exponentially.

That seems to me to be obfuscating the factors that everyone knows. Cases are growing exponentially in some states, while deaths fell tremendously in the states that previously had the most cases. California and Florida didn't have deaths fall 70%; they had deaths more or less plateau, while the age of the infected shifted younger, and now the increase in cases is getting to the point where it more than compensates for the lower death rate.

3 weeks ago there was a blip at the end of the cases chart just like this one. You don't need to be a super genius to understand that the growth rate in deaths is roughly going to lag the growth rate in cases by about the amount of time it takes someone to die, which is about 3 weeks.

3 weeks from today we will have about 2200 deaths per day, based on the 7 day moving average.

3 weeks from today we’ll likely be at around 1000 deaths per day, not 2200. https://covid19-projections.com/us

The case fatality rate for the virus is going down due to increased testing (identifying more cases) and better treatment (Remdesivir, dexamethasone, and better medical knowledge in general).

So it won’t be as bad as a naive extrapolation might make things look. Still bad though.

C19pro is the best model out there, but it only uses deaths as its input. That's good because cases are a noisier measurement, but it's bad because of the lag effect I just described. It will take time to correct if there is a large transient in R.

All else being equal, if the case count triples, the death count is going to triple. Now in this case there are two things that aren't equal. #1 is that the cases are more biased towards younger people, and #2 is that positivity has increased. I expect that these two factors will roughly cancel each other out and the death count will be about 2200. Would you care to place a wager on the over/under at 1600?

The US wide totals are muddling the picture substantially. Deaths have continued to shrink substantially in previously severely affected states like NY or NJ.

The other big issue of course is lag. If you look at daily positive tests, current hospitalizations and newly reported deaths in one graph the lag becomes more visible.

Here I e.g. just used a log y axis to show all three in one plot: https://anarazel.de/t/2020-07-13/all_in_one.png

which imo makes that more visible.