Falcon 9 (spacex's other rocket) wasn't human rated at first either.
The point of starship is to reduce the cost of kg to orbit, by being a fully, and rapidly reusable launch system.
The other long term and loftier) goal is to enable Mars colonization, a mission who's current main blocker is cost of kg to orbit.
By reducing the cost of putting things to orbit, you can do a whole lot more. Starlink is a good example, but if starship works it will be a paradigm shift that will result in a whole new space economy.
But falcon 9 is using a time tested approach and has safety systems. The starship has no backup if for some reason the engines fail to start…?
(For the passenger angle)
I don’t think the limiter to mars is cost to orbit:
- Having a vessel where people can live for a couple of years
- finding someone willing to take the (most likely) one-way ticket to mars
- all the challenges of having a mars habitat. Radiation, dust, etc
Reducing the cost: sure, starship has only been launched with no payload so far so the numbers are yet to be determined… and it’s only impressive (theoretical) numbers are to LEO.
The cost to orbit is the current limiter for a mars mission because no one will invest in solving all the other challenges until cost to orbit is solved. It's also a lot easier to solve the problems you raised when cost to orbit is lowered.
Early variants of starship will not be human rated. That will only happen once Starship has a proven track record. The is also no reason a human rated starship variant could not be built using the same safety systems seen with the dragon capsule.
It sounds like you are having trouble seeing merit in starship. Falcon 9, whilst great, is not going to the end of launch system development. SpaceX believes Starship will bring significant improvements/benefits. This process is no different to how Automobiles and aircraft have seen improvements to their capabilities over the years.
Given the fact that this at the core relies on the `rayon` and `wide` libraries, which are decently baseline optimized but quite a bit away from what llama.cpp can do when being specialized on such a specific use-case, I think the speed is about what I would expect.
So yeah, I think there is a lot of room for optimization, and the only reason one would use this today is if they want to have a "simple" implementation that doesn't have any C/C++ dependencies for build tooling reasons.
Your point is valid when it comes to rayon (I don't know much about wide) being inherently slower than custom optimization, but from what I've seen I suspect rayon isn't even the bottleneck in terms of performance, there's some decent margin of improvement (I'd expect at least double the throughput) without even doing arcane stuff.
It wasn’t tried at scale is what’s being argued. Maos china was not really communist even though they called themselves the communist party. Maybe something like totalitarian state capitalists
Oh, I get that. I’m saying, it barely works at even small scale, like 100 people (there are innumerable failed communes in the US), let alone a small state of a few million. I don’t think the Chinese under Mao implemented communism in any meaningful way, but that’s not proof of communism being unworkable. Its manifest failure on any scale beyond a large family is why I think it’s just not workable for society.
This is the ideal, but it's often false in meaningful ways. In several US elections, for example, we've seen audio leaked of politicians promising policies to their donors that would be embarrassing if widely publicly known by the electorate.
This suggests that politicians and donors sometimes collude to deliberately misrepresent their views to the public in order to secure election.
worse.. a first-hand quote from inside a California Senate committee hearing chamber.. "Don't speak it if you can nod, and don't nod if you can wink" .. translated, that means that in a contentious situation with others in the room, if allies can signal without speaking the words out loud, that is better.. and if the signal can be hidden, better still.
This is an old saying in politics and you're misinterpreting it - it's not about signaling to allies, it's about avoiding being held to any particular positions.
You're also missing the first half, "don't write if you can speak, don't speak if you can nod, and don't nod if you can wink." The point is not to commit to anything if you don't have to.
Also remember cyclists actually have to work to keep their momentum. Did you stop and check for crushed bottles? Glass will puncture your tires on a bike
The problem is memory bandwith. There is a reason Apple Macbooks do relatively well with LLMs it’s not that the GPU is any better than zen5, but 4,5,6x memory bandwidth is huge (80ish gb/s vs 400gb/s)
