That's also true of the predictions of superintelligence, and the models behind them. The core of the essay is actually just extending the data bottlenecks argument that the rationalists have already conceded, to software and math:
https://www.lesswrong.com/posts/icykbWYmdLFYxgvqY/data-bottl...The logic behind the intelligence explosions models is also not based on evidence, so extending and demonstrating flaws in that logic does not require evidence.