As I see it — and keeping in the Wittgensteinian vein — The difficulty we have here is that the term 'knowledge' is vaguely defined across a number of language games, and it's rarely clear which language game we're playing when we invoke it. That causes confusion.
So allow me to go ahead and deconstruct this topic, to see where we end up. When we talk about 'knowledge,' we generally want knowledge to express 'truth.' This is the rationale behind the 'justified true beliefs' paradigm. But 'truth' is a problematic concept. 'Truth' (with a capital 'T', meaning the strongest version of the concept) is something close to a Platonic form: universal, a-temporal, irrevocable, and irreducible. 'Truth' in
this abstract sense is a matter of metaphysics that we have no direct access to. We can presume that the Truth 'is out there' with proper X-Files sensibilities, but we will inevitably Mulder and Scully ourselves trying to get a handle on it.
For example, if I claim that the following statement is 'True':
What I mean is that in any time, place, or context — e.g., the modern US, ancient China, 25th century France, even on an alien planet in a different universe — if we have one of something and a different one of something, and we put them together, we will have a two somethings. But then, of course, I have to realize that while this equation may always be 'True' within the mathematical domain of arithmetic, not everything in the universe is subject to the rules of arithmetic. For instance, if we have one container of water and another container of water and we pour them together, we still only have one container of water (now containing twice the volume). If we have one apple and one orange and we put them together, we do not have two of anything (unless I switch conceptual frames and start talking about fruit).
The point here isn't to quibble with the nature of arithmetic; the point is that 'Truths' are generally only 'true' within bounded domains. We can say that 1+1=2 is a 'truth' so long as we understand that it is true for a particular type of thing: countable, indivisible, immutable objects of a uniform type. If we understand the boundaries, then we can say the claim is 'true', and we have something we can call 'knowledge.'
This is the case even for ridiculous claims. For instance, if I say:
"Purple-striped unicorns are superior to pink-speckled unicorns"
no one would dignify calling that 'knowledge' unless there were a particular context — say a board game or child's TV show — which provides boundaries for that claim. If there's (say) a children's TV show called 'Ultimate Unicorns' in which the purple-striped unicorn shoots a laser out of its horn while the pink-speckled unicorn sneezes up healing mucus, then my claim has truth-value within that context, and we can have a fiery, meaningful debate about whether lasers are 'superior' to healing mucus.
But notice how the nature of 'truth' has changed here. Truth is no longer 'universal, a-temporal, irrevocable, and irreducible' but exists only within a frame of reference (be it arithmetic objects or a particular TV show). And these particular frames of reference happen to be well-delimited. I can specify which objects are subject to arithmetic and which are not; I can specify that we are speaking about a particular show. Can we do the same for other contexts? Can we specify the boundary conditions for physics, climate science, ethics, aesthetics? Even physics clearly stops working at certain point — event horizons, the beginning of the universe, at the quantum level — but the exact boundaries of applicability are still something of a mystery.
Without precise conceptual boundaries, the notion of 'truth' starts to fall apart. Either we make the leap and assert a Platonic ideal of 'Truth,' or we are forced back to mere justified belief.
So now if we can go back to the main point, we can tease apart knowledge and truth, seeing that 'knowledge' has at best an asymptotic relationship to metaphysical 'Truth'. Then we no longer have to use 'knowledge' in the exacting sense of the term — meaning we no longer have to make Platonic assumptions — and merely need to recognize the relationship between claims and boundary conditions that produces practical knowledge.