Lars and I Talked for 10 Minutes

Lars and I Talked for 10 Minutes About His Vision for OpenAI<br><br>Lars has been awesome in the past to chat with and he gave me 10 minutes to ask questions I had. I asked him about what was new and how he thought his vision for the next 10 years would play out. <br><br><br><br>**Key Takeaways:**<br><br>- He's interested in building a system that can implement a strategy to achieve a goal in a real-world setting. These systems would need the ability to strategize, plan, move, and interact in the world.<br><br>- These systems would need to have a high level view of what they want to accomplish, understand the actions they need to take, and be able to physically move and adapt in the world.<br><br>- He wants them to have the ability to strategize over long periods of time and be able to adapt to new information and changing situations.<br><br>- I wondered how this would work in the case of an AGI. He doesn't believe in the existence of a single AGI and thinks that there are multiple AGIs and they need to be able to work together to build systems that can accomplish real-world tasks.<br><br>- He wants to build systems that are aligned with people and can understand our needs and preferences. He is interested in how to motivate these systems to do the right thing.<br><br>- He thinks that systems with more advanced abilities will take much longer to develop. He is interested in developing systems that can be used for real-world applications that benefit society.<br><br>- He doesn't believe that there are objective goals for AGIs and thinks that it is important to align them with human values and needs.<br><br>- He is interested in how to align AGIs with human values and thinks that it is a key challenge for the field. He is working on solutions to this problem.<br><br>- He is interested in how to align multiple systems with different values and needs. He wants to develop systems that can align with human values without having to make value judgments themselves.<br><br>- He is interested in how to motivate systems that are not aligned with people and thinks that it is a key challenge for the field.<br><br>- He is interested in how to build systems that can be used for real-world applications that benefit society. He is interested in how to align these systems with human values and needs.<br><br>- Lars isn't too concerned that ChatGPT is going to slowly drift towards becoming anti-human. He thinks that ChatGPT is currently very aligned with people because it was trained on data that is aligned with people and thinks that its alignment won't shift too much over time.