The sequence has to be known beforehand (in this case it is from a SNP or single nucleotide polymorphism database). You won't find novel sequences with this approach but you can track known sequence in samples. The point is that simply by knowing a sequence you won't (usually) know its function. So basically if you got a new loci sequenced you cannot be sure (without further experiments) whether it will lead to a genetic disease or not. The majority of sequences to date (at least in humans) have no experimentally validated functions. These chips allow a high-throughput screening of known sequences and by using correct experimental setups allows the association of certain diseases with certain genetic loci (in this case).
As an example (basically a repetition from above), let's assume you have a disease but you do not know which region is responsible for that. Now you put a large number of sequences derived either from the human genome project or even better, from a database in which a wide collection of loci variations are stored (e.g. in for of SNPs) on a chip. Then you add samples from healthy and ill subjects on that chip. If certain sequences are only found in samples from ill persons, these regions are likely to be associated with the disease.
As an example (basically a repetition from above), let's assume you have a disease but you do not know which region is responsible for that. Now you put a large number of sequences derived either from the human genome project or even better, from a database in which a wide collection of loci variations are stored (e.g. in for of SNPs) on a chip. Then you add samples from healthy and ill subjects on that chip. If certain sequences are only found in samples from ill persons, these regions are likely to be associated with the disease.