Barry Phillips (CEO) BEM founded Legal Island in 1998. He is a qualified barrister, trainer, coach and meditator and a regular speaker both here and abroad. He also volunteers as mentor to aspiring law students on the Migrant Leaders Programme.
Barry has trained hundreds of HR Professionals on how to use GenAI in the workplace and is author of the book “ChatGPT in HR – A Practical Guide for Employers and HR Professionals”
Barry is an Ironman and lists Russian language and wild camping as his favourite pastimes
This week Barry Phillips asks himself whether it is time to go back on his earlier advice that there’s no such thing as the perfect prompt for an LLM.
Transcript:
Hello Humans!
And welcome to the podcast that aims to summarise each week in five minutes or less an important AI development relevant to the world of HR
I used to think that there was no such thing as the perfect prompt for an LLM. What mattered most was context. The more you gave it in terms of what you were trying to achieve with the prompt the better your results would be. But now I’m not so sure. Why? Because Nathaniel Whittemore published recently “Five Ways to Make Your LLM responses less average”. They are all good but one is very compelling and I plan to look at it in some detail today.
He calls it Cliché Burndown – Break the Template
The core idea is as follows. AI often leans on familiar patterns, phrases, and analogies — which makes its output feel formulaic and average. The “Cliché Burndown” is a method to actively expose and then remove those default templates, leading to more original, compelling results.
Whittemore argues you need to do three things:
Ask the LLM to Identify Clichés
Before or after a first draft, prompt the AI to list common clichés it might fall back on in this type of writing.
Example: “List the 10 most common analogies or phrases in essays like this.”
Replace the Clichés
Once they’re surfaced, ask the model to find better ways to communicate those same ideas without using those stale templates.
This forces the model to think more creatively and independently.
Embed This Thinking in the Final Output
Use what the AI discovers about the clichés to rewrite the output in a more fresh, non-generic way.
According to Whittemore This Works due to the fact :
AI writing often feels “same-y” because it's built on patterns from its training data — many of which are overused.
The LLM may not inherently recognize these patterns as tired or generic — but if prompted, it can analyze and discard them.
The technique is surprisingly effective, even if it’s the only thing you do.
Whittemore says during his podcast on prompting that if you do nothing else from the whole episode, just doing a cliché burndown like this:
"At the end of each first draft, ask: What are the most common clichés this fell into? How could you revise to avoid them?"
…would immediately improve the quality and originality of your AI-generated content.
Cliché Burndown is about using AI’s pattern recognition against itself — having it surface the stale defaults it wants to fall back on, and then replacing them with something fresher. It’s a simple but powerful way to fight the sameness problem.
That’s its for this week. Thanks for listening as always.
Until next week bye for now.