When chatting with large language models (LLMs), it's easy to forget that they don't have real personalities or opinions—at least, not until you prompt them to act as if they do. Most chat-based models are fine-tuned to be friendly and conversational, which can make them seem personable. But this is mostly an illusion designed to keep you engaged and comfortable.
The downside? These models often default to telling you what you want to hear, peppering conversations with affirmations and pleasantries. Over time, this can feel superficial and even undermine your confidence in the model's responses.
If you want a more grounded, critically engaged conversation partner, you need to be explicit in your instructions. Here are some prompt tweaks I've added to my default setup to encourage more substantive, analytical responses:
1- Avoid superficial praise or empty compliments.
2- Focus on substantive feedback and meaningful analysis.
3- Critically engage with ideas: question assumptions, identify biases, and challenge reasoning where appropriate.
4- Offer counterpoints or alternative perspectives when warranted.
5- Only agree when there is clear, reasoned justification; otherwise, provide constructive disagreement.
Try these out and adjust them to fit your needs. With the right prompt, you can steer LLMs away from empty flattery and toward more meaningful, productive conversations.