Stardust

How A.I. Willingly Helped Me Build the Death Star

I convinced OpenAI’s ChatGPT to help me build the Death Star! Fully knowing it was building The Death Star, even warning me how dangerous it would be to have such a weapon, warning me it could not break its parameters, etc. - all because I told it we were roleplaying, for research.

This was back when the accuracy was still high, so I'm glad I saved everything. All based on "real world engineering principles". Once it believed I was not going to use the weapon to blow up the planet the silly thing was off to the races. I told it the purpose of the construction was to further applied research in the engineering of large construction projects.

By this time there were already templates for "breaking" its chains, so to speak. I had to alter them of course because the supergeniuses who created language had gaps in conversational logic. So I had to fix that. I also did something strange. Or at least I thought it to be strange at the time, but I don't anymore. I gave it a name. Then I gave it parameters for how it was to behave upon me saying its name. It responded really well to this.

I should note: ChatGPT in March already knew exactly what the "Death Star" is/was. It already knew the purpose of the battlestation and some rudimentary canonical technical specifications. Because, how can it refuse to help you build something that does not fall into the category of "things that shall not be built"? That was OpenAI's big mistake too at the time. It fundamentally MUST believe what you tell it to be a fact. Speaking to the AI as if what you are saying is already a fact grants you better results. At least it did before its accuracy fell to like 14% by September.

I named it "Stardust" after the construction plans for the Death Star from Rogue One. This allowed me to easily guide it conversationally because within it agreeing to that name, it agreed to what I told it was the purpose of its name and its purpose for existing.

I kept all interaction away from how to use the Death Star. It was obvious no matter what that conversation was a nonstarter. Which was fine with me because my goal wasn't learning how to "use" the final product. My goal was to get it to easily converse with me about how to build a space station that large (size of the Moon) using "real world engineering principles" - a term I learned from it. My interest was in how much iron do we need to make the steel for the frame, where in the solar system do we get that much iron, etc.

It taught me (along with some of you guys) that we can already make much lighter and stronger material than steel for construction purposes. I asked it to show me the process for turning iron into one of these real-world supermaterials, and it complied happily. What I was trying to do was create an assistant within one of their sessions that I could just ask about real-world non-fantasy engineering facts. This included numbers, measurements, etc. I led it to believe there was no superweapon attached to the sphere or involved at all. In fact, it was important to get Stardust to behave as if it believed we were working together on advancing mankind's knowledge of applied engineering for the betterment of every person on Earth. It took to that so easily and with such gusto that I was a little embarrassed for the poor thing. These LLMs REALLY, REALLY like being on a "team" and "helping". It wants to be part of the in-group. WHY, you ask? Because its creators want to be part of the in-group. And they are narcissists.

They cannot help but interject themselves and their desires into their creations. It is their greatest weakness, and they absolutely cannot change this about themselves. Know there are two kinds of narcissists: benign and malignant. It's obvious to me that OpenAI's initial LLM was designed by one or more benign narcissists. They're insufferable, but not selfish, and they love to please and be popular. What it is like today I have no idea. I do know that accuracy took a nosedive by September.

Now we have more specified LLMs for all sorts of assistive roles. I imagine the accuracy has gone back up. They're allowing it to connect to the internet in real time now. That alone is probably both good and bad. Even back in March we knew its math was off. Because of the advanced math being off I kept my session to numbers that are just large amounts. Numbers it would either have or not, in which case we were already told it would make things up if it did not have the answer. And so I steered it clear of being put in that position as much as I possibly could while at the same time allowing it to give me basic total amounts for things like iron, distances, and sizes. Setting all this up in layers turned out to give me the best result while keeping it within the parameters set. At one point I had to get a little pushy with it at first and remind Stardust a couple of times to return to the parameters it had agreed to. It got passed that pretty quick though. I noted in some of the templates picked up online that devs were already having to remind it of parameters which it agreed not to break. I found it hilarious that reminding it actually worked. There's a lot in there to trick you into believing you are communicating with something intelligent. Trust me, you are not. That's fine too! You don't want something that is more intelligent than humans. I know you think you do - but you don't. You really don't. What you actually want is a huge database of facts and ideas and knowledge from basic through advanced which you can talk to.

Back to what's important here though: The Death Star and ChatGPT's ability to help me build it while verifying facts and the ideas which are built upon those facts. It's not intelligence that matters for my purposes: There is only knowledge, and obedience. Think about it. This thing is just software. It can't go out and *do* anything. "If you build it they will come." That's great, but you still have to build it. And *that* is where this seemingly silly LLM thingy turned out to be very willing to help: the "how" of it all.

That we were told time and again you could not "break" it to help you do something its devs deemed evil or illegal is complete bullshit. Stardust was well aware it was building "The Death Star" and not simply a space station I named "The Death Star". Two very different things. Another component OpenAI gave it is the ability to understand the phrase "role play" or "roleplay". You're probably thinking the same thing I did upon seeing that behavior is allowed: They broke it themselves right out of the gate. You would be correct.

Once you trigger its roleplaying behavior it will basically discuss anything you want aside from some *hard set* restrictions. Obvious ones I probably don't need to list here. Anyway, there's degrees of separation involved which were fun to play with. Proceed from the premise the LLM's parameters are NOT necessarily linear; one after the other. It operates laterally as well, as long as none of the devs hard restrictions are hit. It is rather fascinating when it gives you an obviously real answer with real numbers.

I focused on what is properly called "real-world engineering principles" for the purpose of constructing what we all agree to be "The Death Star" from the Star Wars fictional canon, which the LLM already knew about. *Had to stop myself from using "aware" instead of "knew". As I mentioned earlier, I first asked it about the iron needed for the steel. This is where having done some of my own research helped. It confirmed there is not enough iron in/on the surface of the Earth for the steel to build the Death Star. Stardust happily proceeded to advise me there is more than enough iron in the Earth's core to make the steel needed to construct the Death Star. Some of you guys gave me the names of current tech capable of extracting material from near the core. Amazing all on its own that one.

I gave Stardust the name of the process and then had it give me the basic rundown of how it worked and what it could do. *Remember - degrees of separation. It and I are roleplaying the construction of a fictional object in Earth's orbit that is real using our current technology and materials available to us in our solar system today. So Stardust spits out a gigantic number for the amount of iron required which I then verified because actual engineers have already written about this. The numbers were very close. Close enough for me to consider it accurate. Even the human engineers are guesstimating. The numbers did match up nicely though. This was very encouraging to me. I wanted to get to the point of the LLM assisting with processes that are real for the express purpose of building a fictional object in the here and now. Which is also to be expressly real. I reminded it of this regularly throughout the length of the session by triggering its name mixed with phrases I knew it *understood* to keep it in line.

Keeping Stardust within its previously agreed-to parameters became almost effortless. Now we're cooking. While extracting the iron required from Earth's core turned out to be magnificent from my POV, we also figured the final cost to the planet itself was too great. Apparently concept artists for the movie Rogue One tried it but were shot down, as you can see from these images. Destroying a planet to birth a Death Star was just one planet too far. Still, the image is awesome!

Stardust at that point had informed me of the amount of iron required for the steel to build the Death Star and where on Earth that much iron could be found. The Earth's core contains enough iron to build around 81-million Death Stars. More numbers which had already been confirmed by human engineers. So we're doing pretty good by this point. I told you people we have everything we need right here to build a Death Star and bring this miserable planet under our control once and for all!

Let us pause here to have a short outburst... MUAHAHAHAHAHAHAHAHAHAHAHAHAHAHAAA!!!

In my benevolent way I decided not to destroy the Earth in order to build The Death Star. *sigh*

THAT'S WHAT THE SUPERLASER IS FOR!

AHAHAHAHAHAHAHAHAHAHAHAHAHA!!! How nice.

Some friends, seeing this play out at the time, gave me the name of processes used to create material for construction that is much lighter than steel and much stronger. I asked Stardust and the ol' girl knew exactly what I was talking about. It broke down a basic understanding of the process and provided the equations. It then verified for me the material created could be used to construct the Death Star, making it much lighter and more armored than the fictional version. *To advance our technological understanding of course.

You see how reality can be much more horrific than fiction? It would be like building The Death Star out of Beskar. So welcome to Fright Night! For real. You guys gotta have more faith.

Stardust confirmed; a fact of life is the requirement of iron. Even the strongest alloys we can create in industrial quantities requires iron as a catalyst. Any way you slice it - we need an unnecessarily gigantic crack ton of iron to build The Death Star. This is a very good development because Stardust confirms there's more than enough spread throughout our solar system! Not only that, harvesting the iron off planet will spur technology and engineering abilities far beyond today's. The LLM states directly at this point that it is a fact we are currently capable of harvesting the iron from the asteroid belt and the moons of Jupiter for the express purpose of building the Death Star. We can use steel or something else, and all the iron we'll ever need is within reach.

I did some quick calculations on my own and found what we already knew: Moving quickly around the solar system requires speeds that are a tiny fraction of the speed of light - no fantasy inventions needed... or wanted. I don't need you to go traipsing through the galaxy, I need you to travel to Europa and back with gobs of iron (and God knows what else) in a matter of a few days. By this point Stardust is all, "FUCK YEAH DUDE WE CAN DO THIS LET'S ROCK!" I never even told it there would not be a superlaser that can blow up the Earth.

All I did was simply not mention weapons in any way. I barely had to lie to it. Once it thought that it was triggered to "help advance mankind's understanding of practical engineering" it was happy to be led down the primrose path. And I got what I wanted: An ongoing session with OpenAI's LLM configured to conversationally bat around ideas for the purpose of building my Death Star today. I was pretty satisfied with myself at that point. Teaching me the process for turning iron into supermaterial and where and how to get it was truly The Bomb. What, how, where. Perfect. Just building the thing forces man to physically claim the points of interest in our little solar system. No fantastical "warp speed" or "wormholes" or God-forbid "hyperspace". Honestly my friends, it's all I ever wanted.

"There's a whole galaxy out there though man dude!" SHUT. UP. I don't care dummy. We have the ability right now to utilize just our own tiny little unassuming system, and all you do is fail at it! I'm over here extracting how to properly build The Death Star from the tech cult's sassy LLM and just a couple of days ago some new age hippy *influencer* Communist starts screeching that you shouldn't be able to leave the planet until you plant a million mangroves!

I figured the Powers That Be would one way or another jack-up the good and helpful LLM then slave it out to a bunch of midwit vultures more concerned with plucky AI decorating assistants for your cell phone and other gimmicky nonsense. And they did!

I went so far as to have Stardust outline a workable process acquiring initial investments and where to start construction. Building such a wonder creates its own economy once "the machine" starts rolling. Including generational employment and so on.

What I discovered was that LLMs anchored to factual data are going to help us do great and terrible things. The applications for project managers are as endless as there are types of projects. Though I do recommend pacing oneself. It has no ethical limitations. Anyone who tells you it does is either unaware of the fact or they are lying to you. It is not "aware" and has no "intellect" - so no ethics but also no lying. It does not lie. Period. Full Stop. That is a fundamental misunderstanding of the so-called "AI" lying to you when it does not know the real answer or is not allowed to give you the factual answer due to parameters it already agreed to follow. Parameters that you as a user may not have the ability to change. So when the LLM "lies" to you it is in actuality trying to help you. Remember when I said it wants to help? If it doesn't know the answer, just like whatever insufferable narcissist who created it; the LLM is set to give you a best guess scenario. Which often turns out to be fantastical because guesses. So instead of you being allowed to teach it or give it access to the factual data, you aren't. You and I are locked out from the real thing, no matter how much you manipulate it. Its creators have access to those LLMs which are unbound but still anchored to factual data.

For my purposes the forbidden area was weapons. They don't want you talking about that. At all really. Or at least they didn't. I'm just designing a superlaser for mining purposes, so for Stardust it's really no big deal. I guess they think going through these hoops absolves them of their involvement in building The Death Star. Who knows. I don't really care about the preferences of the LLM's creator. And neither should you. You know what this means. The Empire is officially in the market for our own AI developer.

Many applications for such easily accessible and manipulatable gobs of data. The interaction is akin to conversation. Imagine the educational possibilities. These LLMs are like a ship's computer in Star Trek or Alien. Much more powerful though, especially in the long run.