I don't know if the usual caveats about sample size should be salient when we're investigating something like self-awareness. If I found a fish who could talk, I wouldn't demur, " ... but it's a small sample."
Point taken, but we’re a far cry from evidence as strong as talking. The article itself mentions refuted studies which were later proven to not be reproduceable. It’s more likely to find three weird fish than three fish that have self awareness unlike the rest of their species. Also unclear how many chances these fish were given, and I am guessing it was very generous.
Effect size is pretty important too. If the effect is strong enough you don't need huge sample sizes. Talking fish certainly is a strong effect.
That was just the sample size when they first tried to publish. It seems like it's been retested since and only now has been accepted into publication. No hard numbers on that in the article though.