Every time a LLM is fine-tuned it gets stupider and less capable compared to the bare model. openai's legal and social ass covering attempts to neuter their model's output via fine tuning have done the same.
https://arxiv.org/pdf/2203.02155.pdf Page 56/68 in table 14 it looks like the things the fine tunes beat the base models at are basically HellaSwag and the ones that use human evaluations. Otherwise base gpt models are winning.
And just before that they're discussing how the fine-tunes do cause performance regressions on page 55.
I just object to the blanket statement that fine tuning will make a model dumber. It will mostly mean the model is not as good at the original training task, but that really doesn't mean it is "dumber" by any definition.
The question then is what fine tuning does to tasks that are neither the original training task or the fine tuning task. This will depend on the fine tuning task. It seems that instruction fine tuning improves performance on tasks that involve human interaction, so I have a hard time seeing it as the model becoming dumber. Other fine tuning tasks, such as removing toxicity, may have a higher cost on unrelated tasks, so there one could say they caused the model to become dumber.
It is unclear why this may be, but to speculate, it may be due to classifying large regions of the distribution as 'off limits' and therefore less structure about these regions is modeled in detail (since it is now not needed, and summed up as 'as a LLM, I cannot'). It is certainly strange that making a model less deplorable will make it worse at coding in general, but it does appear to be a real observation.