Notes

• Major things this decade
  • AGI
  • Finding life off Earth
    • Drake equation - We know that there’s gazillions that have the potential to host life if life got there, but we don’t know how many on which life independently arose, and we don’t know that probability
  • Decoding animal communications
• New ways to think
  • telescopes have now got to the stage on the Earth-bound telescopes as well as the JWST are big enough to do atmospheric spectra on planets
  • AI helps pattern match everything that’s going on
  • global sensing revolution
    • early days still - less than 1% into a big future and AI is accelerating that
    • sensing side - Planet now images the entire land mass of the Earth once per day at about 10, 30, 11 a.m. local time (180 doves/super doves)
  • if you want to get a quick understanding of something, take a picture. If you want to scientifically measure it, take a spectra.
  • spectra = fingerprint of objects on the earth
  • real-time accounting system of what’s happening on Earth (a common picture of what’s happening)
  • Data companies = winner-take-most sort of situation (e.g. Bloomberg)
  • Mainstream LLMs are now multimodal - beyond text/language (audio, visual imagery, video)
    • you can automatically search for things without having to have trained any model
    • estimate land use change over time, plot some pie graphs of the changes
    • count new roads or buildings
  • AI companies are going to have to proactively steer that technology. Otherwise, their default path will be to accelerate our blowing up of the planetary boundaries
  • space debris
    • relatively simple things can ensure that we don’t cause a problem, like keeping the satellites lower. where the atmosphere of the Earth just self-cleans those orbits
    • The bigger problem is not satellites, but the debris, which are the leftover bits of satellites when satellites blow up and things.
    • (1) ban anti-satellite technology - 10,000 pieces of of debris that you can’t even track because it’s so small and that could be catastrophic to collide with other satellites
    • (2) mega constellations need to stay low
    • (3) traffic management system and debris removal services
      • laser scheme for nudging space debris
      • the known pieces of debris if two are going to collide you nudge one using light pressure so the light just nudges it so that it misses the other piece and then it stops the cascade with one laser system in one telescope in the uh pole you need in antarctica or the arctic
        • NASA doesn’t know what to do with projects that are under a billion dollars. So this was only going to cost 10 or 20 million
        • Australians that have been testing a laser through a telescope system there to nudge debris
    • satellite build $$$ –> $?
      • 300 million per spacecraft to do that in the Earth systems budget
      • After R&D, we can pump out that spacecraft for about six or seven million, including launch
  • sending cell phones up in the space - Consumer electronics would work in space and we’d be able to do spacecraft this 1,000x-ish cheaper
    • which we still haven’t seen the full implications of
• Inspiring things
  • Rainbow Mansion
  • Cosmic Duty vs Cosmic Tragedy (we have wiped out 70% of life by numbers, by population decline. It’s 80% for wild mammals. So there’s 80% less deer and moose, and there’s 70% less fish in freshwater rivers and lakes. There’s 75% less insects than there were 40 years ago. There’s half the corals have gone, half the forests have gone.)
  • Jevons paradox - the default way in which AI is going to have an impact in our capitalist-driven society is definitely going to be accelerating the exploitation of the planet
  • Copernican events - nudging us out from being the center of the universe on everything
• Planet
  • eight spectral bands at three meter resolution, all in optical and near infrared
  • Terra Bella acquisition from Google - 50 centimeter resolution
  • hyperspectral satellite - 420 color bands, just like our eye has three colors. It’s 420 colors spanning a wide spectrum. So it’s starting at 400 nanometers out to 2.5 microns
  • PPP w/ NASA JPL and a consortium of players to develop the most sensitive hyperspectral instrument in the galaxy
  • We estimate the market to be 50 to 100 billion a year just for our earth imaging data and analytics sitting on top of it
  • convolutional neural nets - bespoke trainer model
  • boomerangers - They go to SpaceX, they go to Google, and then they realize that it’s better at Planet.
  • dirty infrastructure off earth less pressure on our ecosystems (Gerard K. O’Neill)
• EO Tracking
  • the type of species of tree or coral reef or grass or what have you on the ground
  • The shade of paint so well for defense and intelligence applications that they can say, this tank… was made at that factory because their shade of green or whatever it is camouflage is slightly different from the other factory
  • detect gas emissions so um turns out methane gas for example absorbs a particular uh set of uh uh of light in a very narrow uh um spectra And so by looking for that and the ratio between that and other spectral bands nearby, we can actually tell the volume of methane in the column between the satellite and the ground
    • measure the amount of methane gas leaking from a facility like a gas leak or cows farting or from a landfill site or whatever
  • watch construction projects taking place
  • crops are healthy
  • precision agriculture - automatically extract out your field boundaries of your different fields, automatically say which crop type, is it corn or soy, and automatically figure out which area of the farmer’s field your field is doing less well, needs water, needs fertilizer, needs help in some way
  • count objects - building detection, road detection, airplane detection, ship detection
• Scale
  • detect where the deforestation is in 8 million square kilometers of the Brazilian Amazon
    • value prop: halved the deforestation rate - they reduced deforestation rates by more than half, 3,000 interventions last year, confiscating $3 billion worth of assets
  • battle damage assessment near Ukarine - monitoring 18 million square kilometers every day
    • value prop: half of them were being shot down and not coming back. So instead, now they just look in our imagery and if there’s no changes they don’t need to send that drone saving them billions of dollars
• Metrics
  • price performance of satellites
    • cost reduction: 1,000x over 4x for the rocket
  • launch cost per kilogram
• TODOs (look into)
  • convolutional neural nets - bespoke trainer model
  • Australians that have been testing a laser through a telescope system there to nudge debris

Transcript: EP 43 Will Marshall <> Ashlee Vance on Core Memory

Source: URL