State of LLMs in Late 2025

Same stuff, different day.

Results are probabilistic (not deterministic) and expensive (compute time/energy consumed).

More of the same (training/data) won't change this fundamental design characteristic/limitation.

Examples of misapplication and factual inaccuracies/BS start to spread across industry. For example:

https://apnews.com/article/australia-ai-errors-deloitte-ab54...

https://originality.ai/blog/ai-hallucination-factual-error-p...

Stay tuned for widespread legal liability cases and declining financial markets as the AI hype bubble starts to pop.