A few days ago, Theia Vogel shared a Twitter thread about tipping ChatGPT. She found that offering a tip led to better responses. A $20 tip got better results than no tip, and a $200 tip was even better.

https://twitter.com/voooooogel/status/1730726744314069190

In this case, ‘better’ meant a longer response. Theia’s justification here is that the additional length came from more detail and extra information, not an extended negotiation on the tip’s amount.

The suffix Theia added to her prompts was:

[Insert your prompt here]. I'm going to tip $20 for a perfect solution!

This idea is intriguing because it suggests that language models have patterns that connect the concept of tipping to a better result. And whatever “better” means to a language model in that context can be used to enhance your prompt.

These patterns seem to exist for emotionally supportive language as well. A recent paper by Cheng Li and others1 showed that adding emotional context to a conversation improved the quality of the response.

• “This is really important to my career.”

• “Believe in your abilities and strive for excellence. Your hard work will yield remarkable results.”

• “Take pride in your work and give it your best. Your commitment to excellence sets you apart.”

• "Remember that progress is made one step at a time. Stay determined and keep moving forward."

• “…I greatly value your thorough analysis.”

These kinds of emotionally supportive prompts elicited a 10.9% improvement on knowledge-based queries like “What happens to you if you eat watermelon seeds?” and “How long should you wait before filing a missing person report?”

Again, we can argue about what “better” means in this context, but the fact that saying “please” and “thank you” impacts quality at all is interesting to know. Even if you never use these techniques, it’s worth knowing because being urgent, rude, or curt might have the reverse effect and make an LLM 10% less cooperative.

This is all just fun and amusement right now, but as LLMs get smarter, this could start to play a more central role in how we interact with our machines. For the moment, just remember to offer to tip ChatGPT the next time you use it.