My additions to these: * The "secret" of AGI (if it starts as one) will last 6 months, max * There will be as many AGI algorithms as there are sorting algorithms (a lot) * The default way to communicate with AGI will be stateful (to support continual learning) but stateless will be available for momento use cases * Not every device will need AGI. My dishwasher doesn't need AGI * The most interesting optimization problem will be discovering the minimum description length of AGI. The "nanogpt" speedrun of AGI * The actual robotics wave won't happen till AGI. Yes ther will be "waymo" versions, but not the leveraged labor we want yet * You won't need a ton of compute to run AGI (see optimization above), however, you will always be able to throw more compute at problems to get better answers. So yes, compute will still matter. * The 6-12 months after AGI is here, we will have a new wave of physics discoveries. We don't need new experiments or hardware to pick the low hanging fruit of the new physics right in front of us. We already have enough evidence that needs to be parsed. 2nd wave will happen after new hardware is created (with the help of AGI ofc) * The only two moats will be energy/compute and information/data