First Question: In many problems where we find probabilities by counting, we produce a sample space with $N$ equally likely outcomes. If for $S$ of these outcomes we have "success," then the probability of success is $S/N$.
In this problem, there are two natural sample spaces of equiprobable outcomes.
(a) Imagine choosing $4$ bulbs and putting them in a bag. There are $\binom{24}{4}$ different possible bags, all equally likely. How many bags have $4$ bad bulbs? It is clear that there is only $1$ such bag. (There is good reason to write instead that there are $\binom{4}{4}$ such bags.) So the probability of "success," if you can call $4$ bad bulbs success, is
$$\frac{1}{\binom{24}{4}}.$$
(b) Imagine picking out a bulb, then another, and so on, and putting them down in a row, in the order in which they were selected. Then there are $(24)(23)(22)(21)$ possible such rows, all equally likely. How many possible rows are there with $4$ bad bulbs? Clearly $4!$. So the desired probability is
$$\frac{4!}{(24)(23)(22)(21)}.$$
The mistake in the proposed solution is that the denominator is the one we get if we use bags, while the numerator is the one we get if we use rows. The first step in the analysis is to decide on an appropriate sample space. All counting of outcomes refers to that space.
Second Question: Again, there are several ways to get a sample space of equiprobable outcomes, and therefore several ways of getting at the answer.
(a) Imagine that before the performance, someone chooses a pair of seats "at random," and puts Reserved signs on these seats. There are $\binom{n}{2}$ ways of choosing $2$ seats, all equally likely.
How many of these sets of two seats are adjacent? The leftmost one of the two adjacent seats can be any one of seats $1, 2, 3,\dots,n-1$. So there are $n-1$ adjacent pairs, and our probability is
$$\frac{n-1}{\binom{n}{2}}.$$
But $\binom{n}{2}=(n)(n-1)/2$, so our probability simplifies to $\frac{2}{n}$.
(b) Forget about theater seats, let's line up $n$ people in a row at random. We ask for the probability that A and B are next to each other. The $n$ people can be lined up in a row in $n!$ different ways, all equally likely. Put A and B in a bag (it's OK, they like to be close). Now we have $n-1$ abstract people, really $n-2$ people and a bag. These $n-1$ objects can be lined up in $19!$ ways. Cut the bag open. Then A can go to the immediate left of B, or to the immediate right, for a total of $2(n-1)!$ arrangements in which A and B are next to each other. Note that all arrangements with A and B next to each other are uniquely obtainable in this way. Thus our probability is
$$\frac{2(n-1)!}{n!},$$
which again simplifies to $\frac{2}{n}$.
(c) For a change of pace, we work more directly with probabilities. Imagine that the theater is empty, and A comes in and chooses a seat at random. Then B comes in and chooses a seat at random. We calculate the probability that A and B end up next to each other. This can happen in two ways: (i) A sits at an end and B sits next to A and (ii) A sits at a non-end, and B sits next to A.
The probability that A chooses an end seat is $\frac{2}{n}$. Given that A is in an end seat, the probability that B sits next to A is $\frac{1}{n-1}$, since only $1$ of the remaining $n-1$ seats work.
The probability that A chooses a non-end seat is $\frac{n-2}{n}$. Given that A has chosen a non-end seat, the probability that B sits next to A is $\frac{2}{n-1}$, since there are $2$ seats that work.
Thus the probability that A and B end up next to each other is
$$\frac{2}{n}\cdot\frac{1}{n-1}+\frac{n-2}{n}\cdot\frac{2}{n-1}.$$
The above expression simplifies to $\frac{2}{n}$.
Comment: The three solutions to the second problem are not complicated, but in each solution the very simple answer $\frac{2}{n}$ is obtained by algebraically simplifying a more complicated expression. There ought to be an argument that yields $\frac{2}{n}$ directly, but I have not come up with one.
There are $n-k+1$ possible locations for $k$ people occupying adjacent seats, and there are $\binom{n}k$ possible locations for $k$ people, so the probability in the first question is $$\frac{n-k+1}{\binom{n}k}=\frac{(n-k+1)k!(n-k)!}{n!}=\frac{(n-k+1)!k!}{n!}\;,$$ as you say.
In the second question there are still $\binom{n}k$ possible choices of $k$ seats, but there are now $n$ of them that have the $k$ people in adjacent seats, so the probability is
$$\frac{n}{\binom{n}k}=\frac{nk!(n-k)!}{n!}=\frac{k!(n-k)!}{(n-1)!}\;.$$
(I’m assuming that the seats in the circle are individually identifiable, i.e., that seatings that differ by a rotation are still different seatings.)
Best Answer
Sasha has given the standard answer. Here is a different approach using the fact that each arrangement is equally likely, but which avoids factorials or conjoined people.
Seat A first. Then the probability that B is seated immediately on A's right is $\dfrac1{n-1}$ (since there are $n-1$ people who are not A) multiplied by the probability there is in fact a seat to A's right $\dfrac{n-1}{n}$ (since if A sits on the far right there is no seat to the right), which is $\dfrac{1}{n}.$
Similarly the probability that B is seated immediately on A's left is $\dfrac1{n}$. So the probability they are sitting together is $\dfrac2{n}$.
If the people were sitting at a round table, it should be obvious the answer would be $\frac2{n-1}$ when $n \gt 2$.