Across 30 different tests carried out by the UTSA researchers, 440,445 of 2.23 million code samples they generated in Python and JavaScript using LLM models referenced hallucinated packages. Of the ...