One example – gradualism in evolution.
It was clear to me in logic and statistics that large populations would change slowly, but small populations could change rapidly.
It was also clear just from looking at the moon through binoculars, that big things hit it, leaving big scars, so similar things probably happened here on earth (but life and erosion etc covered the tracks so to speak).
Those two things alone suggested clearly that we would find a vast spectrum of mechanisms where major changes occurred over relatively short times and/or in relatively restricted places.
A modern synthesis is vastly more complex than the accepted theoretical structures I was initially taught, and rejected as being too simplistic.
We now understand that many different mechanisms (not simply the two outlined above) can alter the rate and direction of evolution in some contexts, and that such things can be recursively selected for (our immune systems being a prime example).
A second example is the notion of truth.
I was taught, and most still believe, that one may know truth.
That now seems highly improbable.
We now have sufficient information about the complexities of reality, and the mechanisms of our own sensory and neural networks systems, that I am clear that my experiential reality is a subconsciously created model of reality, and the best models I can have can only ever be simplistic approximations to the reality that seems to actually be present. Reality does in fact seem to be that complex.
Some of those model can be very useful, and accurate to quite a few decimal places in some contexts, but that does not mean that they are true in any absolute sense of the word, and it does make them very useful approximations to whatever reality is (which is all that is really relevant in evolutionary terms).
When I was young, most accepted that science delivered Truth.
Now I, and I suspect a majority of serious scientists, accept that reality seems to be sufficiently complex and contain sufficient levels of fundamental uncertainty, that the best any of us can hope for is contextually useful probabilities, not any sort of hard knowledge.
And in some common contexts, those probabilities can be a very close approximation to unity (to 10 or more decimal places, so near enough for most practical purposes, but not actually unity).