It's slightly misleading to say that there were better yields for 5nm than 7nm because according to the graph in the article the yields have been almost exactly the same, maybe slightly lower. The article's assertion that the defect rate is a key metric seems dubious since it seems to have been roughly the same for the last three generations of chips and thus kind of a nonfactor. But in fact low yield rates were the reason Intel famously had to delay its 7nm processes. The fact that defect rates have been kept roughly constant seems to suggest they can go even further into 3nm processes, but this is the problem with extrapolating from graphs with too little data - at 3nm they'd run into very challenging physical limits and it's not clear we'll ever get a 3nm chip.
It is amazing to me that we can achieve such low defect rates on projects so complicated as highly brain-insecure sticks of meat that evolved to hunt and forage. Under the principle of radical skepticism the true defect rates could be much higher.
It sounds like you are conflating "Defect Density" and "Yield". These are not the same. Yield is a function of Defect Density. The function is exponential. Even the appearingly "slight" improvement shown in the Defect Density shown in graph is significant in terms of Yield.
It is amazing to me that we can achieve such low defect rates on projects so complicated as highly brain-insecure sticks of meat that evolved to hunt and forage. Under the principle of radical skepticism the true defect rates could be much higher.