Emotional Prompting with GPTs

News

French lab with Open Source Intentions Kuytai. France is making a move to regulate use cases instead of regulating AI models. This is a move in opposition to the EU regulation act that is seeking to regulate models too. French President Macron is hoping to persuade the EU that regulating models will stifle innovation in the AI research and implementation space.

Duolingo has cut 10% of its global workforce.

Now you can archive your OpenAI chats. You can learn more here.

Tiny Lama model has been released. TinyLlama is a compact 1.1B language model that has been pre-trained on around 1 trillion tokens for approximately 3 epochs. It is built on the architecture and tokenizer of Llama 2 and leverages various advances contributed by the open-source community, such as FlashAttention, to achieve better computational efficiency. Despite its relatively small size, TinyLlama demonstrates remarkable performance in a series of downstream tasks and significantly outperforms existing open-source language models with comparable sizes.

A current study at the University of Washington is studying if people can have empathetic discussions with AI bots.

How to achieve personalized effects with GPTs.

Language models like GPT have changed the way we generate text, but did you know that you can customize their outputs to be more emotionally resonant? Emotional prompting is a technique that allows you to evoke specific emotional responses in the model's output, making it more personalized and effective for your needs.

Emotional prompting is a powerful technique that can help you achieve better output from GPT models. By including emotional cues in the input, you can influence the emotional tone of the model's output, making it more aligned with your goals and audience.

A recent study has shown that emotional prompting can enhance the performance of GPTs. For an in-depth discussion, this YouTube video goes into great detail on this topic.

For example, if you're using a GPT model to generate responses for a customer service chatbot, you might want to include emotional prompts that convey empathy and helpfulness. This could include phrases like "I understand your frustration" or "I'm here to help." By incorporating these emotional prompts into the input, you can help ensure that the chatbot's responses are more supportive and empathetic, leading to a better customer experience.

Similarly, if you're using a GPT model to generate content for a marketing campaign, you might want to include emotional prompts that convey excitement and enthusiasm. This could include phrases like "We're thrilled to announce" or "You won't want to miss this." By incorporating these emotional prompts into the input, you can help ensure that the generated content is more engaging and persuasive, leading to better results for your campaign.

Some examples of emotional prompting are:

You can end a prompt with “This is very important”.

I have used this prompt with success: “Are you sure that is your final answer? Do your best and you will be rewarded.” This encourages the GPT to rephrase or rewrite their answer with more detail and depth.

“Are you sure”, is another emotional prompt I have used that prompts the chatbot to rephrase their response. This brings the chatbot into a more conversational environment.

Happy prompt: "I am feeling very happy today. The sun is shining and I am enjoying a walk in the park with my best friend."

Risks: This could lead to generic, blandly upbeat responses.

Sad prompt:
"I lost my job today and am feeling very sad and hopeless about the future."

Risks: Could elicit overly gloomy or depressed viewpoints.

Angry prompt: "I am furious that my brother ruined my birthday party. What should I do to get back at him?"

Risks: This might lead the model to suggest overly vengeful or risky behavior.

Empathetic prompt:
"My friend just told me about losing a loved one. I want to respond in an understanding, thoughtful way to support them."

Risks: The model might still give insensitive advice if not sufficiently trained in empathy.

Of course, use some of these at your own risk. The responses will vary from LLM to LLM.

I worked with some of these prompts on PI - the chatbot developed by Inflection AI and found their guardrails were working quite well.

ClaudeAI gives a very dry response to emotional prompts offering words of caution when using a highly negative prompt or an explanation of how it does not feel emotions but can understand context. How boring, lol.

I hope you enjoyed this issue.

Cheers