EVE Online And Gemini: It's All About The Prompts

Since the announcement of the managerial buyout of Fenris Creations from Pearl Abyss I've switched my AI program of choice from Copilot to Gemini and the entire suite of DeepMind applications. After about two months I thought I'd do a little review of how well the products of Fenris' financial and research product perform for me when used on EVE Online. 

My newest and oldest characters having a beer

Before heading straight into Gemini I want to take a quick look at Google Flow. The problem I have is coming up with backgrounds for the characters. That is kind of a given since the stars of the game aren't the players' avatars but the ships they fly. I'll need to see what I can do with ship images.

Output from my global PLEX market scripts

I am not going to dive into the assistance I received in upgrading my EVE Python scripts. For now my scripts focus on analyzing the global PLEX market. I used Gemini to clean up the code from my initial efforts using Copilot plus extending the data base my scripts can use. The API I access every day only holds up to 13 months of data. A couple of weekends ago I used Gemini to write data files locally I can access for historical data. I already had the code written in another script but all I had to do is copy my old code into Gemini and I had new code. Yes, I still had to troubleshoot problems because AI is not perfect, but the process was a lot less painful than what I experienced with Copilot.

A Sacrilege fit created by Gemini

Now for the function I found the most fun: ship fitting and theorycrafting. If players use Gemini (or any AI, really) poorly a lot of nice, shiny, EXPENSIVE kill mails will show up on zKillboard. Until AI gets a lot better, fitting is going to feel a lot like manually figuring out how to fit a ship. As the title of this post says, "It's all about the prompts."

I went through the fitting process with a Sacrilege. Tell Gemini the ship being fitted, the type of content

to engage with (PvE vs PvP), why are you running the content, and the location in New Eden (Minmatar Republic vs Caldari State, for example). Gemini might ask some clarifying questions but eventually will provide a fitting.

This is where the double checking begins. Use the in-game fitting tool to make sure of things like do you have the skills to fly the fit and are all the fitting slots being used. For example, in the image above my fit is intentionally leaving one of the high slots empty.

Then comes what I consider the fun part. Have Gemini explain the fit to me. One of the most important question (or prompts) to ask is why Gemini decided not to take advantage of a ship role or bonus. Sometime doing so improves the fit. Sometimes the reply explains something I didn't consider. In other words, I learn, or at least remember, things about EVE I wouldn't have by fitting the ship unassisted.

The Minmatar Ship Tree

The next two uses are more for some blogging projects I have. One is the actual number of spaceships in EVE Online. Looking at the ships listed in EVE's many ship trees I came up with 374. I then came up with a crazy idea: how many ships has Fenris introduced to EVE while Star Citizen has been in development. Not exactly an easy question to answer since Cloud Imperium has spent close to 14 years developing SC. So I made a spreadsheet with the names of the ships, ship details, but perhaps most importantly, when the ship was released.

Remember the title of this post? My life would have been a lot easier if I had used efficient prompts at the beginning. I wound up using the ship tree to identify the 100+ ships that weren't present in the initial data set. I spent a couple of hours going through finding all the missing ships, duplicates, and misspellings. But I do have the answer to the question of how many ships that are on the ship tree have the developers introduced since Star Citizen was announced: 133.

The last project I used Gemini for was related to my EVE ship spreadsheet. The spreadsheet currently has a column indicating if a ship was introduced before or after the introductory announcement of Star Citizen. But what about the eras of EVE? I wanted to divide the game into eras but I wasn't sure if I wanted to just use "First Decade", "Second Decade" and "Third Decade". 

I came up with a pretty good initial prompt with different options. The three eras I came up with, using Gemini's research to back up the decision, were:

• Era A: The Classical Sandbox & Accumulation (2003 – Summer 2013)
• Era B: Intended Fixes & Unintended Consequences (Late 2013 – Late 2022)
• Era C: Reconciliation, Ecosystem Design, & Independence (Late 2022 – Present)

Something strange happened during this last project. I began thinking of Gemini as a partner and not just an application to use. I know AI programmers try to achieve that goal, but my prompts became a conversation, bouncing ideas off Gemini. My little idea to create eras I can use to think of EVE history became a fairly detailed outline, lining up in-game developments with real world business decisions.

I decided to let Gemini write a brief article on the history of EVE. I then acted as editor to remove all the inaccuracies I could find. The brief history turned into 2500 words and 6 1/2 pages in Google Docs as putting in the corrections I found added an additional 2-3 pages to the story.

For those wondering what a Gemini-assisted story looks like, below is the title and link.

The Three Epochs of New Eden: A Macroeconomic History of EVE Online

I'm not quite sure about some of the null sec political stuff. Yes, I was around but never really played in null sec due to my allergy to bubbles. But as much as I've gone over the Google Doc I think, except for the length, the article is pretty good. While I may not formally publish the piece I am going to use the Google Doc as background information when writing about EVE Online in the future.

I've used Google DeepMind products, particularly Gemini, quite a bit since the MBO of Fenris Creations from Pearl Abyss brought the AI giant into EVE's orbit. One thing I've learned is that AI is not a magical wand one can wave to make things easier. The less work one puts into using the tools, the worse the results. The one thing I've become convinced of is that experience makes the use of AI more effective. One has to have the ability to recognize errors in order to fix them.