🥢🧵 Kojima’s Sticks and Ropes
Hideo Kojima once said that in most games, "you see a lot of hand-holding. You see tutorials telling you how to use the controller and how to progress through the game. My philosophy is to give players a rope, not a stick. Give them something to help them, but also something to tie onto and swing over a chasm with." This philosophy of empowerment and agency can be applied to large language models (LLMs) as well.
"The stick is the first tool that mankind created to put distance between himself and bad things - to protect himself," Kojima said. "The second tool mankind created is a rope. A rope is a tool used to secure things that are important to you. Most of your tools in action games are sticks. You punch or you shoot or you kick. The communication is always through these sticks. I want people to be connected not through sticks, but through what would be the equivalent of ropes."
As we look to the future of LLMs, we must ask ourselves: are we simply using these models to generate content, answer questions, and summarize information? Or are we truly empowering users to think critically, write effectively, and make connections we otherwise would not have uncovered? In this post, I’ll explore the concept of empowerment as it pertains to LLMs, and examine two toy examples of how these models can be used to encourage active engagement and creative thinking.
Empowering Users with LLMs
My main critique of LLM apps today is that they are often used for question answering, content generation, and summarization, which takes away the agency of critical thinking. In contrast, providing guidance and support to users on how to think, organize, and write more effectively can empower them to become more active and engaged learners, thinkers, and communicators.
Using LLMs to guide users in their writing and thinking process can be compared to providing them with a rope. Instead of relying solely on generated content, users should be encouraged to develop their own unique writing style and thought processes while the LLM provides guidance and feedback along the way.
In the past month, I've noticed many question-answer bots using Langchain and GPTIndex. They extract data, search it, and build question-answering systems. This reminded me of my experience as a senior engineer. When junior engineers asked me simple questions, I had to decide whether to give them the answer or teach them to solve problems themselves. I knew that if I always gave them the answer, they wouldn't learn and grow.
Examples of Empowerment
To illustrate how LLMs can be used to empower users, here are two toy examples examples:
Study Notes app
- A study app that simply summarizes large amounts of information into predefined study guides.
- App that not only summarizes information, but also generates open-ended and personalized questions for the user to try to answer.
- Provide suggestions for where the answers might be found while generating personalized content
- Finds connections and analogous to subjects of interest for the user based on known competencies.
- A journaling app that simply allows users to transcribe voice memos and cleans them up for later reference.
- Encourages the expansion of ideas through thought-provoking questions, weekly check-ins, and feedback.
- This helps users improve their thinking and communication skills by expanding their own ideas.
- Asks questions in a way that encourages accountability, challenges, and offers support to people's ideas.
Large language models have immense potential to transform the way we learn, think, and communicate. However, the current meta of LLM apps tends to be geared towards passive consumption rather than active engagement. By shifting the focus towards empowering users with LLMs, we can encourage critical thinking and creativity while providing guidance and feedback. The examples of the Study Notes and Journaling apps demonstrate how LLMs can be used to give users agency and enable them to become active learners and thinkers. Let us embrace the philosophy of empowerment and agency in our approach to LLMs and harness their full potential to transform the way we communicate and learn.