There's an argument that LLMs and other deep-learning systems will remain limited and not generalizable to new domains.

Unfortunately, the things they argue these systems can't do soon are not relevant to 90% of jobs, and are probably less than 25% of the remaining 10% of jobs.