[hackernoteng]

Models get smaller and perform better, requiring less GPU hardware.

The "hallucination" problem persists.

More prompt programming tools are created for AutoGPT-type task processing.

Lots of companies spend lots of money to roll out LLM-based apps but most fail because user adoption and rejection due to poor performance/hallucinations.

The successful solutions will be around assisting content creators and programmers (co-pilot, journalism, graphic design, etc.)

Will be built into internet search (google/bing) and office tools (word, excel) but won't be the primary way work gets done.