“Historical arguments against AI progress, such as models' inability to handle semantics or paragraph-level coherence, have consistently been overcome by finding workarounds or through continued scaling.”