Obviously stuff like LLMs produces much more impressive results as of now, that's a given. OTOH who knows - neural networks have also had a long-ish period when OCR seemed to be the pinnacle of what they can deliver before they exploded via Deep Learning/Transformers/LLMs and what not.
Eurisko demonstrated superhuman abilities to play strategy games in early 1980-th, and even used strategies from VLSI place-and-route task in planning fleet placement in games. This is knowledge transfer between tasks.
The bogon flux must NOT be interrupted.