Have LLMs Become Too Predictable? At the Expense of Creativity.

Not even 1.5 years ago, The NY Times did an interview with the “first prompt engineer”, a job with a “salary range of between [USD] 280,000 and 375,000”. The Internet was suddenly littered with courses, or organigrams for the less invested, teaching us the dos and don’ts of creating prompts. I took one myself. Much of what I learned is now outdated. Not unlike the prompt engineering career path itself, which is quickly being replaced by a prompt tuner, i.e. the LLM itself. Something we all should’ve seen coming. The student has surpassed the master. Except…

Three LLMs, Three Predictable Articles

I recently asked ChatGPT 3.5, ChatGPT 4*, and Gemini to write an article about my five-day trek to Annapurna Base Camp. They all received the same prompt. What they spat out in return was a grammatically (mostly) well crafted and extremely poetic account of any mountaineering adventure. The repeat occurrence of specific words and phrases — like “sense of peace”, “a roller coaster ride of emotions”, “a journey of self-discovery”, “stairway to heaven”, and some variation on finding my “resilience” — was also no surprise. This is an example of an LLM behaving exactly as it should. If you want some deeper insight, have fun working your way through this article.

Since AI is not capable of independent creative thought, all the content LLMs generate is simply an amalgam of information they have already been fed. Just regurgitated in a slightly different order after…