“Weather models can be useful, because we can validate them against real a thousand times a year (six hour outputs daily), and find their strengths and weaknesses.”
Since modeling weather and climate (which I think of as basically weather integrated over time and space) are so complex, one should expect a fair degree of error that can then be used to provide information useful in tweaking the equations used in those models.
And those errors should be all over the place, in all directions. One key troubling feature of the climate models is that they all run “hot” - if you look at graphs of the predictions of global temperatures over time, starting, say, a couple of decades back, just about all (one Russian group excepted) predict temperatures higher - and often much higher - than what we actually observed. When the errors all give results that run in the same direction (and the direction that’s favored politically) that suggests they’re not random.
What you are observing is the bias that results from groupthink. It is politically fashionable (and rewarding) to see "global warming" as the boogieman. So the modelers build it into their models as the "correct" vision of the future. They do this by the "tweaking" the author mentions.
If your model shows global cooling, no grants for you! You are likely to lose your job, as several examples show!