Some organization publishes a dataset yearly, under CC BY 4.0, some other organization now aggregates this data into a single dataset, which they post on their github repository that is under the GPL 3.0 license, it the license of this aggregated dataset CC BY 4.0 or GPL 3.0?
It depends on a number of factors:
- Is the dataset subject to copyright or database right protection at all? In Europe, the general answer is "yes, it probably is protected by sui generis database rights, and maybe copyright," but in the US, this is dependent on the degree of originality in the selection, organization, or presentation of the dataset. Even if the US does recognize a valid copyright, it only attaches to those creative elements, and not to the actual data itself (see Feist v. Rural, and especially the rather lengthy obiter dicta the Supreme Court wrote in that opinion). If the data is not subject to any protection in a given jurisdiction, then the choice of license is irrelevant to consumers operating in that jurisdiction.
- Did the aggregation process involve creating or computing any new data, or did they "just dump all the data in one table" and call it a day? In the latter case, database rights may still attach to the resulting dataset, but they are irrelevant as consumers could simply get the data from the original source under CC BY, and aggregate the data themselves. However, if the aggregation process actually created new data, then it may be subject to its own independent copyright and/or database right, which is licensed under the GPL according to your question.
- Is the dataset accompanied by any software or other material? If so, that presumably has its own copyright status to consider, and is subject to whatever license it is originally distributed under.
The GPL is probably not relevant here, unless the relevant copyright holder(s) contributed the data set and specifically signed a CLA or changed the license for the contribution.
The CC BY-SA is the one worth talking about. It might or might not actually apply to some portion of the data. Some of the data might or might not be copyrightable. The compilation (selected set) of the data might or might not be copyrightable. Does the data include software? Does the data include images? Do you have model releases for any pepole in the images? Are you in France, where Moral Rights might apply? Does the CC BY-SA licensor even have any of the underlying rights in the data at all?
Data licenses might be better for handling these issues.
But yeah, the GPL probably isn't relevant.
This is not legal advice and I am not your attorney.