Farzad: Agents Will Use Compute Orders of Magnitude More Than Humans + A New Profession Called Creator + SpaceX Seeks Up to 456 Starship Launches a Year + Gemini Argon Plays Cornhole

Farzad said Tuesday midday CT that the strain on GPUs/CPUs from agents will be BEYOND massive, then quoted that post with a Tesla/SpaceX compute call. Tuesday night Farzad posted on a new profession called Creator and on the Golden Age of Mathematics. Farzad also liked and RTd SawyerMerritt on SpaceX's Starbase launch request and RTd posts on Micron.

"Orders of magnitude more than humans ever could"

Farzad, Oct 13 midday CT: My head spins thinking about what agents will be able to do by the end of next year. Their memory is going to get exponentially better. So will their ability to do long-running, super complex tasks.

Farzad wrote that the strain on GPUs/CPUs is going to be BEYOND massive as agents become extremely good at squeezing every bit of compute out there. In Farzad's view, distributed compute is basically mandatory at this rate, because agents are going to become unbelievably good at squeezing every bit of available compute globally. Farzad added that centralization of compute will still be super important and needed, but distributed compute is going to be unbelievably valuable too.

Farzad's close: If there was a way for me to tap into everyone's idle CPU/GPU capacity on their desktops, I would do it in a heart beat.

About 26 minutes later, Farzad quoted that post. Farzad wrote that it's inevitable Tesla/SpaceX get into the GPU/CPU/Memory game with Terafab or whatever the next-gen of computing is. Basically anything that requires silicon to run compute.

Farzad's premise: If one assumes digital AI can be 100x more effective per watt of compute than a human with that same compute, then the demand for compute will be beyond anything that has ever existed in the history of humanity.

Farzad gave a personal example: a video Farzad created with Grok Bot last week, fully end to end and automated, that Farzad says is performing extremely well. To render it, Farzad could have used the CPU/GPUs available in Grok's computer, which Farzad wrote is not that many. The other option was local compute on a Studio & mini, which Farzad wrote would make it go WAY faster.

Farzad: With agents, in theory, I am only limited by amazing ideas for content, and everything else will be handled by the agents, including the rendering. So in a perfect world, I will have my local machines rendering videos 24/7.

Farzad added that the more compute Farzad has, the more videos Farzad can get done, plus whatever other use cases the CPUs/GPUs can be leveraged for. Farzad called it only ONE example of probably millions.

Farzad's TL;DR: Agents will use CPUs/GPUs/Memory orders of magnitude more than humans ever could. Farzad expects that to grow exponentially as models and agents get better at executing complex tasks.

Both posts are Farzad's projections, not reported news. Tesla/SpaceX getting into GPU/CPU/Memory is what Farzad expects. Neither post points to an announcement.

Check agents and compute here. Tesla/SpaceX compute QT here.

"A new type of profession: Creator"

Farzad, Oct 13 late evening CT, quoted benjitaylor. benjitaylor, Oct 13 late morning CT: I no longer see any compelling reason to believe that taste is some permanent boundary LLMs won't cross. It's already happening.

Farzad: This is going to enable some absolutely insane shit. Imagine not having to worry about the nuances of what makes something great. You just spit the idea out and agents will take care of it for you.

Farzad thinks this inevitably leads to a new type of profession: Creator. Not a content creator per se, but just a Creator. As Farzad describes them, these are people that have harnessed AI agents, both physical and digital, to bring their ideas to life in their fullest form.

Farzad's guess: The most ambitious & most thoughtful Creators will probably be some of the wealthiest people on the planet, and/or solar system?

Farzad wrote that the question becomes who controls the agents. Do we control them? Does anyone? Farzad's close: The future is starting to look CRAZY.

Farzad also liked benjitaylor's post. Farzad also liked Elon's reply to it, Oct 13 late afternoon CT: True.

The Creator profession is Farzad's projection. The point about taste is benjitaylor's.

Check Creator QT here. benjitaylor here. Elon here.

"The Golden Age of Mathematics"

Farzad, Oct 13 late evening CT: We're entering the Golden Age of Mathematics. Humans are now able to deploy AI to solve the infinite problems in Mathematics. Humans have approximately solved 0.00% of the Mathematics that exist in our reality.

The Golden Age call and the 0.00% figure are Farzad's own view, not a measured number.

Check Golden Age of Mathematics here.

SpaceX asks the FAA for up to 456 Starship launches a year

Farzad liked and RTd SawyerMerritt, Oct 13 midday CT. Sawyer: SpaceX is officially seeking FAA approval for up to 456 Starship launches and 912 landings per year from Starbase, a big increase from the 25 launches and 50 landings allowed today.

Sawyer laid out two phases. Phase 1 would include up to 100 annual launches and up to 200 annual landings, 100 of the Starship and 100 of the Super Heavy. Launches and landings could occur at any day or time of the week. Sawyer quoted the FAA: "As the vehicle continues to mature and reliability improves, launch and landing rates are expected to increase. Starship and Super Heavy are designed to require minimal post-flight refurbishment."

Phase 2 would include up to 456 launches originating from, and 912 annual landings at, the Boca Chica Launch Site at Starbase, Texas. Launches could occur at any time of day, but only Monday through Friday. Sawyer quoted the FAA again: "This phase represents a mature program launching more than 100 times annually and would be implemented within several years."

Per Sawyer, public comments are due by Nov. 30, 2026.

This is a request for approval, not an approval. The numbers and FAA quotes are as Sawyer posted them.

Check SawyerMerritt RT here. SawyerMerritt here.

Gemini Argon teaches a humanoid cornhole

Farzad liked normandipalo, Oct 13 midday CT, and later quoted it with "Wow!": I'm still half in disbelief. We asked Gemini Argon to teach the Apollo humanoid to play cornhole, and it zero-shotted it.

normandipalo: All we gave it was a video of a person doing it. According to the post, Gemini Argon then took control of the robot, generated its own sim env, performed some sysID and trained launches on the real robot. It took a few real shots to improve and then landed many consecutive airmails.

The post says the process was so long and articulated that they asked the robot itself to explain what it did, which is shown at the end of the video. It even came up with the gestures. The post tags Google DeepMind and Apptronik: Great demo for a great model.

Check normandipalo here.

"Have your bot talk to my bot"

Farzad liked poteto, Oct 13 late evening CT: it won't be long before "have your bot talk to my bot" becomes common!

The post listed some fun use cases of multiplayer grok bot over email: debugging (you can send my bot bug reports!), customer support, booking meetings and events, group buys, and buying and selling stuff off craigslist. poteto also asked: what have you used bot mail for?

Check poteto here.

Micron and the long term AI story

Farzad RTd heyitsyashu, Oct 13 late evening CT. heyitsyashu, Oct 13 evening CT, quoting an Elon post: If you needed more proof how vital $MU is to the long term AI story.

In the same minute, Farzad also RTd Elon. Farzad also liked the same Elon post. Elon, Oct 13 late afternoon CT, replying to RealNickMugalli: Micron has been reasonable regarding future memory prices, while others have been extortionate, which is something today's customers won't forget. Elon called it the smart move for long-term customer loyalty.

RealNickMugalli, Oct 13 late afternoon CT, had posted that JP Morgan just put out a new note that's super bullish on memory and storage. Per RealNickMugalli, $MU now has 26 long term deals covering 35%+ of revenue through 2030, $150B in contracted backlog and $32B in customer cash already committed. RealNickMugalli added that over 75% of next year's output is already sold.

The Micron figures come from RealNickMugalli's summary of the JP Morgan note. The pricing comment is Elon's.

Check heyitsyashu RT here. heyitsyashu here. Elon RT here. Elon here. RealNickMugalli here.

New videos

Farzad posted four new videos, two Oct 13 evening CT and two Oct 14 morning CT.

Check Inference Is Where the Real AI Money Goes here. Until 2015, No Orbital Booster Had Ever Landed Itself here. Nobody's Ready for What Elon, Bezos & Palantir Are Building here. Starship Could Take You to Shanghai in Under an Hour here.