It's like saying "I'm gonna pick a random number between 1 and a trillion", and then picking 999,999,999,995. Probably not a smart idea given that you don't want anyone else to be able to guess your number.
2. the birthday problem basically halves the exponent security wise. The rule of thumb: If you have N possible outcomes, then after around sqrt(N) guesses the probability of a collision approaches 0.5. So, for birthdays, it's 365 outcomes, so with 19 or 20 people your risk of collision already approaches a half. For BTC private keys, there are 2^256 possible, so with 2^128 guesses you'd approach a likely collision. Fortunately, that's still 1e38, so if you check 1e10 per second, you'd still need 1e20 years to get there.
2^(256/2) is way, way bigger than the number of used bitcoin addresses, which is about 33 million according to this csv [1].
[0] https://en.wikipedia.org/wiki/Birthday_attack#Mathematics
