Eli Fast

CS @uscviterbi
Current intern @spacex prev @valaratomics

[Eli Fast @EliFast1]

Compute is what unlocks the rare futures we are after.

[Elon Musk @elonmusk]

@aaronburnett If this is all SpaceX achieves, we will have fallen far short of our goals

[Eli Fast @EliFast1]

What people don't realize is that if AI works it grows GDP like population growth, but since there are no new humans it also grows GDP per capita by an equal amount. This is why this time is different.

[Eli Fast @EliFast1]

@CJHandmer 1. When to have kids
2. How your father and startup founder roles interact

[Elon Musk @elonmusk]

@AJamesMcCarthy Read books, talk to people & iterate rapidly with hardware & software

[Eli Fast @EliFast1]

Economically productive AI must accumulate knowledge like a human during work. Even the best hire in the world is fairly useless on their first day, which is roughly where we're at with AI.

[Eli Fast @EliFast1]

I think orbital datacenters are easier to build than economically productive AI. We really need to get the AI working.

[Eli Fast @EliFast1]

If AI works it will grow the economy like population growth not productivity growth. The days of 2.5% annual GDP growth will be over and a sci fi universe will be real within our lifetime.

[delian @zebulgar]

The best part about this rotation from software into industrialization

Is that venture capital dollars and companies will finally be transforming the physical world

The 2010s deployed billions of dollars, but if you stepped out on the street, the world looked the exact same...

[Elon Musk @elonmusk]

@IterIntellectus Harnessing even a millionth of the Sun’s power, which is extremely difficult, results in an economic value far more than a million times than that of Earth’s current entire economy

[Brad @Brad08414464]

many scientists think humans are nothing more than meat machines. AI is here to show us what a machine really is. you will realize that we are not a machine. we are so much more

[Eli Fast @EliFast1]

The future is going to be vastly bigger than anyone realizes. Feeling so bullish on humanity right now.

[Alex Hoganson @Alex_Hoganson]

startup idea: switch earths rotation direction so we can launch rockets from the west coast

[Elon Musk @elonmusk]

Congratulations @SpaceX team on an epic first Starship V3 launch & landing!

You scored a goal for humanity.

[Eli Fast @EliFast1]

@redbullfuturist Hell yeah

[SpaceX @SpaceX]

First full stack of Starship V3

[SpaceX @SpaceX]

Starship and Super Heavy V3 together at the Starbase launch pad for the first time

[Claude @claudeai]

We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.

This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.

[signüll @signulll]

most people think ideas come from:

- insight
- intelligence
- taste
- reading
- vibes

but in practice they actually come from:

- building the wrong thing
- hitting a constraint
- getting embarrassed by users
- realizing the obvious thing you missed
- noticing the second order effect you couldn’t see from the couch

a really great idea is the *output* of the work, not the input.

[John Carmack @ID_AA_Carmack]

256 Tb/s data rates over 200 km distance have been demonstrated on single mode fiber optic, which works out to 32 GB of data in flight, “stored” in the fiber, with 32 TB/s bandwidth. Neural network inference and training can have deterministic weight reference patterns, so it is amusing to consider a system with no DRAM, and weights continuously streamed into an L2 cache by a recycling fiber loop. The modern equivalent of the ancient mercury echo tube memories. You would need to pipeline a bunch of them to implement modern trillion parameter models, but fiber transmission may have a better growth trajectory than DRAM does today, so it might someday become viable.

Much more practically, you should be able to gang cheap flash memory together to provide almost any read bandwidth you require, as long as it is done a page at a time and pipelined well ahead. That should be viable for inference serving today if flash and accelerator vendors could agree on a high speed interface.