A single dimension of general psychopathology, p, has been hypothesized to represent a general liability that spans multiple types of psychiatric disorders and non-clinical variation in psychiatric symptoms across the lifespan. We conducted genome-wide association analyses of lifetime symptoms of mania, psychosis, irritability in 124,952 to 208,315 individuals from UK Biobank, and then applied Genomic SEM to model the genetic relationships between these psychiatric symptoms and clinically-defined psychiatric disorders (schizophrenia, bipolar disorder, major depressive disorder). Two dimensions of cross-cutting genetic liability emerged: general vulnerability to self-reported symptoms (pself) versus transdiagnostic vulnerability to clinically-diagnosed disease (pclinician). These were only modestly correlated (rg = .344). Multivariate GWAS identified 145 and 11 independent and genome-wide significant loci for pclinician and pself, respectively, and improved polygenic prediction, relative to univariate GWAS, in hold-out samples. Despite the severe impairments in occupational and educational functioning seen in patients with schizophrenia and bipolar disorder, pself showed stronger and more pervasive genetic correlations with facets of socioeconomic disadvantage (educational attainment, income, and neighborhood deprivation), whereas pclinician was more strongly associated with medical disorders unrelated to the brain. Genetic variance in pclinician that was unrelated to general vulnerability to psychiatric symptoms was associated with less socioeconomic disadvantage, suggesting positive selection biases in clinical samples used in psychiatric GWAS. These findings inform criticisms of psychiatric nosology by suggesting that cross-disorder genetic liabilities identified in GWASs of clinician-defined psychiatric disease are relatively distinct from genetic liabilities operating on self-reported symptom variation in the general population.