My Decision
Decision Made
I have made a decision about my company I want to start. As I've said many times, I feel very uncertain about exactly what I want to do with AI. The other day my wife and I went to breakfast and afterwards decided to work on our computers. I was having trouble focusing so I decided to talk to Gemini about my future business. I told Gemini to be a professional business coach and to be harsh, realistic, and to push me. Needless to say, it did not hold back and honestly I was surprised by how much it actually knew about me. We discussed a "safe" option, still related to the restaurant industry and a innovative option, where we are trying something bigger and bolder. When discussing the safe option, it pointed out that I seemed bored just talking about it so what's going to happen when times get tough. It was a very valid point and honestly I was already reluctant about it. After that we decided to "bet it all" and do something new, exciting, and completely outside the restaurant industry.
Burn the Boats
When I said I didn't want to go down the safe path, it said "Good. Burn the boats" assuming that it knew that I was a fan of David Goggins. The next part helped a lot. It broke it down in three parts, "The Pivot", "The Technical Moat", and "The Capital Problem". The pivot was about how LLM's are cool but the real future is world models and to focus on something looking towards the future instead of just doing what's been done. In the next section, it told me that it's great that I'm getting certificates and taking course but real founders build things. I'm not going to "out-math" kids from Stanford so I need to build things from something I understand. More importantly, I need to build things that break because real founders build things that shouldn't work. Then it pointed to looking at a complex system that I understand deeply.
Click
That's when it clicked for me, before I go into the third section, what complex system do I understand? Rock Climbing and skydiving, which is all well and good but what about it? One aspect of myself that I have always been proud of is my ability to handle life or death stress. When it truly mattered, I've never cracked under the pressure.(knocks on wood) I've had two cutaways skydiving, a cutaway is when you have to cut away your main parachute and deploy your reserve. Both instances I felt completely in control and didn't have a moment of hesitation, actually the first one was kind of fun. My wife on the other hand tends to freeze up in scary, dangerous situations, which is a very normal part of life. That's when it hit me, LLM's are really just giant prediction machines and world models are trying to understand how the world works. Well, what if we could determine at what point someone is going to freeze from fear. How many people in this world, not just skydivers, freeze when something scary is happening to them. How many surgeons need to make a split decision to save someone's life and while professionals they can still freeze from fear of the wrong decision. Soldiers can freeze on the battlefield which could cost them or their team's lives. My question is can we use AI to determine if and when someone is going to freeze.
But How?
This hit me like a bag a bricks, because it was something I understood and more importantly I found really exciting. I not only thought that this could be really interesting but could also be really helpful. Imagine if we could not only predict when someone will freeze but also if we could make it so they don't freeze at the wrong moment. Not to gloss over the third section, "The Capital Problem", Gemini pointed out that if I couldn't produce a simplified proof of concept than this would all be for naught. I need to build a specific, high fidelity world model for a very niche thing to start. I will tell you right now I have absolutely ZERO clue as to how to do this. Luckily, I have AI to help. I immediately felt overwhelmed, but Gemini was able to at least point me in the right direction. There is a lot, and I mean a lot of work to do but it does feel nice, and terrifying, to have a heading.
What Now?
My first course of action is to start a deep learning course so I can at least understand how to build a small neural network. I have also been tasked with starting to record data when I go to the climbing gym. Our first big goal is to record of bunch of data of hopefully me getting scared if not frozen while climbing at the gym and just being able to parse the data. To clarify, were not building even a simple neural net to start. just seeing if I can get enough data to organize and build a program to parse it. So, at least, that sounds more doable than they long term goal. For now, I will continue to post probably more philosophy of AI and other things I find interesting and I will update everyone as to the progress as it happens.