DP_DeFi (🦇 , 🔊)

Hyper Simulation and Erebor Google released a tool, called Genie3, that can simulate 3d worlds from prompts. DeepMind CEO, Hassibis noted that this would "allow them to make generative worlds" Previously for training robots, getting real world data was a huge bottleneck. See: self driving, Teslas. Hassibis noted that Genie3 was their answer to robotics training. You can generate infinite worlds where This has large scale military applications, for example using neural nets to make complex aerodynamic models of jet fighters. the NN is just a giant function approximation at its limit unlike AI Safety oriented chatbots, or self driving (don't hit the pedestrian). the comparable for military tech is "he fired a mortar into an area so as long as it's better targeted than that we're probably somewhat okay" eventually you'll move 1 level of abstraction higher for military applications. you already have intense training manuals for pilots that read somewhat like LLM instructions. These will be fed into models, generate worlds, and be trained iteratively so here's a multi layered framework for moving up the AI death machine 1. Vanilla combat - ppl run around with guns 2. Augmented combat - ppl have guns, and computerized planes that do airstrikes, starlink in the field + palantir makes soldiers smarter 3. Drone warfare - human piloted drone. remove the human, all the other intelligence piped directly into the drone which is now a streaming data tool 4. Modular warfare - exactly like 3 except now there are simulations of how you deploy all your drones with a subsequent real time sim of how you're doing vs the battle plan with live adjustments to the plan (this is potentially AI driven) 5. Abstracted Warfare - rather than having a real battle you use Genie 3 to simulate all potential battles, drone paths etc - and deliver an initial plan as well as pre compute nearly every contingency 6. Dynamically Abstracted Warfare - now you assume your adversary also has abstracted warfare so need to simulate their likely model set vs your model set. so now you're simulating their simulator 7. World Model Warfare - bc of 6, suddenly everything becomes a game of who can get the other person's abstraction model. so stealing peoples' synthetic data, training methodologies etc become the primary war method. bc you beat the adversary every time if you have their training method. so now you're modeling the likelihood of different levels of data leakages (your current op sec) around with battle simulation. so basically simulating the people running the simulation and their vulnerabilities 8. Deterministically Modeled Abstraction - adversaries likely have a strong statistical view about who is likely to win a conflict based on simulations -- so can negotiate prior to fighting (this is a weird, optimistic outcome). or, if cornered -- a darker output is that you'd be likely to escalate tactics (nukes, etc). For example if Russia were able to say deterministically via a world model it'd lose prolonged ground warfare ... you get the idea 9. Sociopolitically Modeled Abstraction - because you know many things in advance, the question then becomes simulating responses to various extreme military actions. this would include the public response to escalation, further deployment, etc. so at level 9 - you're now forced to simulate all of society. you already probably were forced to simulate this because you needed to know how bad your security leaks were (how many spies working at google as % etc) So basically the end state of simulated warfare is a nearly limitless expansion of compute -- because 1] it enables near term tactical superiority 2] you can predict casualty rates in advance 3] you have a view of vulnerabilities you can adjust proactively 4] you have algorithmic view of society level costs of increasing deployment levels (something which, especially in the US has been a big reason for war failures over past 50 yrs) 5] entire wars can be avoided or calculated to perfection in advance which means they will be I started thinking about this because I kept asking myself, why is Palmer Luckey, the founder of Anduril (AI Drone Warfare) - starting a crypto bank called Erebor (a LOTR reference). what does that have to do with drone warfare And it makes a lot more sense when you start thinking about the above points. if we're moving to hyper simulated warfare then crypto becomes an important vector to pay intelligence operatives and also to monitor bc hostile operatives will likely be operating here. So you'll want to basically create a military industrial complex style crypto banking system that also works with AI start ups, because eventually if we are at a Full Scale Society War with China- you basically need to ensure you've got your sht locked down Or maybe they just want stablecoin exposure for fun. who knows Regardless, this comes back to the "Big Difference" between the current AI boom and the 1999 internet bubble. Ppl can scream until they're blue in the face about the 'failed enterprise deployments'. but the difference between the internet and AI is the lethality level. Generated worlds mean War Games. Easy to say Palantir is a bubble at 100x sales. But - from a position of humility you have to ask. What if it's not ?

I believe this is where we finally break out of the traditional four year cycle revolving around the Bitcoin halving. Higher for longer. Supercycle.