Were LLMs used much in programming the simulator itself? Safety-critical systems like those should not use computer-generated source.
I have absolutely no idea where you obtained this statement. NASA has been using an automatic code translator from MATLAB to C++ for ages [1]. I think this is pretty much the definition of “computer-generated source”. SpaceX is researching AI use in rockets [2]. Most autonomous cars heavily use LLMs neural networks to drive the cars, which is arguably much more critical than a satellite.
I do know how familiar you are with space engineering, but this is a simulator, not embedded code. The key point here is to make it as close as it can get to reality. The following plot (orange is the simulator and blue is the satellite telemetry) shows that we were spot on:
I was the lead engineer of the attitude and orbit control subsystem of the Amazonia 1, the first remote sensing satellite entirely designed, integrated, and operated by Brazil. I can assure you that your statement about “computer-generated code” is entirely irrelevant. The safety of a modern system is ensured through comprehensive integrated analyses and tests in a verification and validation campaign.
15 Likes
Maybe they use transformers, but I doubt they use LLMs, but probably you intended the former
2 Likes
I meant neural networks! Thanks, I corrected the post.
1 Like
This is not quite correct. For example, code generated by the tools such as Embedded Coder within the Matlab & Simulink ecosystems is used in aerospace industry: https://www.mathworks.com/solutions/embedded-code-generation/flight-code-generation-aerospace-systems.html. Worfklows exist that aim towards compliance with safety standards: https://www.mathworks.com/solutions/functional-safety.html.
4 Likes
I guess MATLAB’s code gen is deterministic, where as LLMs are stochastic. I have nothing against LLM use. Just pointing out the difference.
The simulator dashboard looks amazing…
1 Like
That’s no more “computer-generated” then the assembler produced by a C compiler. As my message made quite clear, the issue is with LLM-produced slop that is lately being heavily promoted on various free software and open-source project forums.
That reply is the very opposite of “reassuring”. Do we need to have satellites to literally start landing on our heads to notice the problem with vibe-coded critical software?
Just use any LLM to be at least a little more educated on the matter because it is literally impossible (physically speaking) for any satellite to “land on our heads" due to software bugs… Again, I have no idea what your background is, but you clearly have no knowledge about how modern systems engineering works. It is not about how a software was coded, it is about testing.
This is a post about using Julia technology to build a satellite simulator visualization using an amazing vibe-coded package by @kahliburke, which provided a huge contribution to the community instead of just hijacking threads for the sake of trolling.
1 Like
@Mikhail_Kagalenko, this is now the second time you’ve coopted a concrete topic about a particular package with a general discussion you clearly want to have about gen AI usage.
Like I said the first time, such a general discussion could be welcome here if done in a respectful manner. You can start it!
What’s not ok is the disrespectful dismissal of others’ opinions. It’s also not okay to derail every discussion about a package that uses gen AI with a discussion about gen AI.
This isn’t about censorship of a particular topic, but rather about keeping topics on-topic and maintaining a respectful and welcoming place to all in accordance with our community standards
7 Likes