I'd expect a model that is expected to be accurate over a handful of decades would in fact be predictive on the scale of a year or a few years.
I also specifically left open the avenue that a model didn't predict this as a likely outcome but did have smaller probability outcomes that would have predicted a temperature decrease.
Its reasonable enough that the model may have found this as unlikely, but unreasonable if the model was so wrong that it wasn't even aware of this as a possible outcome in the first few years of the model. The data simply can't be useful 30 years out when it has such blind spots only a year or a few years out.
For clarity, when I mention models here I mean all recent environmental models attempting to predict global climate or specifically ocean temperatures. If we have no idea why the temperature has dropped then every model failed to provide any scenarios where that would have happened.
If you dig into what those models do, well... cover your eyes. Some frequently run off the rails and predict a climate like Venus or Pluto. They're non-deterministic and don't always do this. Obviously that's due to bugs but the bugs don't get fixed. Instead they just do lots of runs and pick the ones that don't go crazy.
More importantly, because models are iterative the data becomes more and more useless as time goes on and predictions are missed. In this case, any model that expected the Atlantic to increase in temperature this year will likely be even less accurate for next year's prediction. Sure the models can be re-run with the latest climate data, but that just moves the goal posts and raises the question of when we realize the models aren't predictive.