In answer to the second part of your question (an answer to the first part having been more than adequately given by others above): MD5 is considered weak due the proofs of attacks against the cipher (i.e., changes that can be made in the plain-text that do not result in changes in the MD5 sum). Other hashing techniques may not be as easily susceptible to essentially arbitrary hash collisions (at least such arbitrary collisions have not, as yet, been shown to be possible with the SHA-2 set of hashes, etc.), and hence, an attacker is less likely to be able to replicate a hash hashed in a non-MD5 technique (theoretically, of course, hash collision attacks are possible against any hashing function; it would not succeed as a hashing function if this were not the case; the question is how easily an attacker can succeed in "faking" a "correct" plaintext, that is, one that hashes to the same hash value).
Incidentally, the MD5 sum of a plaintext is not necessarily safe because it contains "less" data or is "lossy", but because, from an arbitrary plaintext, it computes a sum-value within a fixed range (for plaintexts < 128 bits, the MD5 sum, in fact, contains more information than the plaintext...), and, therefore a number (theoretically infinite) of plaintext could all align to the same MD5 hash.
