kiratp

Raptor is a breakthrough engine. If NASA with nation state resources and 23B is 6 years late, let’s cut SpaceX some slack for building a profitable launch business with multiple successful vehicles while also being very close with Starship.

> As of 2019, only three full-flow staged combustion rocket engines had ever progressed sufficiently to be tested on test stands; the Soviet Energomash RD-270 project in the 1960s, the US government-funded Aerojet Rocketdyne Integrated powerhead demonstration project in the mid-2000s,[6] and SpaceX's flight capable Raptor engine first test-fired in February 2019.

https://en.wikipedia.org/wiki/Staged_combustion_cycle

If you want to learn more

https://youtu.be/LbH1ZDImaI8

Replace rocket with 747.

It’s hard to be excited about a new 747 from Boeing that flies one way and then you throw it away.

Artemis is not changing the economics of space in any way. Just going back to the moon is not very interesting IMO.

SpaceX is making significant breakthroughs that you can appreciate if you understand the tech.

> As of 2019, only three full-flow staged combustion rocket engines had ever progressed sufficiently to be tested on test stands; the Soviet Energomash RD-270 project in the 1960s, the US government-funded Aerojet Rocketdyne Integrated powerhead demonstration project in the mid-2000s,[6] and SpaceX's flight capable Raptor engine first test-fired in February 2019.

https://en.wikipedia.org/wiki/Staged_combustion_cycle

If you want to learn more

https://youtu.be/LbH1ZDImaI8

Same person, but on YouTube for those staying away from Tiktok

https://youtube.com/c/CeciliaBlomdahl

I've always wondered, what if you were to wire up a panel and one light to the grid, but the then rest of the house is on a separate independent circuit? Technically the house is on-grid...

The good ol’ Paradox of Tolerance: https://en.wikipedia.org/wiki/Paradox_of_tolerance

> The NASA rules for human-rating its own internal-designed craft are way looser than the rules for human-rating commercial craft. SpaceX and Boeing had to jump through way more hoops to get their platforms human-rated by NASA.

Can you share a couple of examples of this? Curious...

I only see one role here that is even ballpark for a SWE in SF.

https://h1bdata.info/index.php?em=&job=software+developer&ci...

The prevailing wage test fails to account for stock and bonuses and is basically a joke.

If we are arguing “it’s a Turing Machine!” Then you could argue that WiFi light bulbs are

> implemented on a very specific set of hardware, restricted in its use by software whose only purpose is to maintain the illusion

Our devs aren't cleaning their machines - it goes to Apple. As long as their techs can do it it matters not. Its cheaper that way anyway.

This design optimizes for the rest of the (365 * 3) - 1 days that the machine takes less space on your desk and is whisper quiet.

I’m not trying to compare the details of SOC performance.

Just that with the same hot caches, the average change-build-test loop that developers do 100+ times a day is just faster on the M1 Max.

The target arch doesn’t matter for the change-build-test loop d vs do. All that matters is how fast can you compile your code to test it.

If the final x86 production build takes longer it doesn’t matter - that happens on the cloud anyway.

Edit: Rust builds are very parallel until linking. No different than any other LLVM build.

Osmos.io is built on Rust so that’s LLVM. No commercial vendor involved.

There is Mold (https://github.com/rui314/mold) as an alternative but it doesn’t support MacOS yet and that’s a blocker for our team.

This is one of the reasons why - pro audio - https://www.youtube.com/watch?v=kIQINCWMd6I

Here is the target market

https://www.youtube.com/watch?v=kIQINCWMd6I

This is Apple's core creative audience.

Exposing my limited understanding of that level of the computing stack - it is but Apple seems to have very very good caching strategies - filesystem and L1/2/3.

https://llvm.org/devmtg/2017-10/slides/Ueyama-lld.pdf

There is a breakdown in those slides discussing what parts of lld are single threaded and hard to parallelize so I suspect single thread performance plays a big role too. I generally observe one core pegged during linking.

We plan to move to it once MacOS support lands (for the laptops).

My workstation has a 3990x.

Our "world" build is slightly faster on my M1 Max.

https://twitter.com/kiratpandya/status/1457438725680480257

The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point. I expect the M1 Ultra to crush the 3990x on compile time.

My workstation has a 3990x.

Our "world" build is slightly faster on my M1 Max.

https://twitter.com/kiratpandya/status/1457438725680480257

The 3990x runs a bit faster on the initial compile stage but the linking is single threaded and the M1 Max catches up at that point. I expect the M1 Ultra to crush the 3990x on compile time.

100% true. We are turning spreadsheets into ETL flows at https://42layers.io

The path to empowering non-developers to program computers necessarily has to involve tools they are already familiar with.

Edit: typo

Or internal tools.

Internal tools should not exclude people needing accessibility features.

I'll try and answer that to an extent here :)

Lots of companies trying to build low-code solutions to help business people glue systems together. However, for pretty much all of these solutions, while the end user isn't writing code, they are forced to think like a programmer - if/else, loops etc.

We are taking a very different approach.

We've built a transform engine that can be trained on transforming data from a source structure to a destination structure using a few (10s) of examples of source and destination. We can do this transformation without falling into the trap of figuring out acceptable confidence levels - a trap that most ML systems fall into, and thus have a hard time with enterprise usage.

We couple that with dynamic, configurable integration infrastructure ("connectors" in old school enterprise speak) that can send+receive data to/from lots of systems over many protocols and serialization systems.

End result is that end users can connect systems together with a few clicks and by providing a few lines fo training examples, not unlike what a business person would give a dev and say "extract a CSV from SAP and put it in that FTP folder. the CSV needs to look like this file"

100% agreed that people are performing higher level functions today that software can't perform.

However, I think your first point about multi-system interconnectivity is ripe for change.

It's been the case that the literal act of running a business has been humans serving as copy-paste bots between systems, both internal and external. Come to think of it, from a purely software point of view, businesses look a lot like giant, multi-system ETL machines, except that the individual steps in the pipeline (Salesforce, SAP, Netsuite, etc.) don't talk to each other. This is even worse when it comes to interactions with other businesses (customers/vendors/partners) - everyone has different systems and none of them talk to each other.

So we fall back to the lowest common denominator - Email + attachments (XLSX, PDF), CSV over FTP etc.

The fundamental problem is not very different from the challenge of human language translations. Getting SAP to talk to Salesforce is a similar class of problem as enabling an English speaker to talk to a Hindi or Mandarin speaker. If the latter is a solvable (solved?) problem, I don't see why software talking to software is that different. There are of course domain specific challenges, like the fact that both systems being translated between require 100% translation accuracy.

We are working on solving this at https://42layers.io. It's early days for us, but this is exactly the problem we are solving.