Coriolis: The Great Dark First Look Review

So after getting frustrated with ChatGPT, I ordered Coriolis: The Great Dark which is published by Free League Publishing. I have purchased DragonBane and read through it extensively and had liked what I saw. Definitely liked the production values, and all the reviews of it were great (to me, it's a 5E-adjacent but streamlined with OSR feel Fantasy Adventure Game, to use the latest terminology I've discovered).

So Coriolis: The Great Dark (CGD) is a 'new version' of Free Leagues' (FL) Coriolis: Third Horizon (CTH). I do not have CTH but it got good reviews also as a more wide open space opera rpg. Both use The Year Zero Engine (YZE) mechanics (dice pool) with some changes for CGD I guess. I had no experience with YZE. The Great Dark is specifically separated from the Third Horizon setting, although it is nominally related. Owners of the Third Horizon have voiced displeasure at that, but it seems either at GM discretion or future releases they could be 'rejoined'. That was zero impact for me just picking up the Great Dark. 

The Great Dark was 'inspired by 19th century expeditions, deep sea diving and pulp archeology.' It is a 'roleplaying game about exploration and intrigue. Join dangerous expeditions, navigate the Byzantine schemes of Ship City, and hunt for the leys that unlock the mystery of this Lost Horizon.' That all sounds great!

From my first partial reading, it looks like there are some issues. One reviewer said it seems like it is an 'environment looking for a system'. Typically you have the game mechanics, then you design a milieux to play in using those mechanics. The Great Dark wanted a remote environment ('frontier') with the suspense of exploring possibly cursed archeological ruins. So it designed mechanics for that, and it seems a little heavy handed. The 'Delve' is the exploration of ancient ruins, which they have structured as 2D vertically to 'get the feel' of spelunking, I think. But it is all mechanics driven, roll for this scan for information, roll for resisting this threat, mark off 'supplies' which cover everything from food to light to oxygen. 

The overly mechanic driven 'delve' just carries forward from the character creation, which railroads characters into the Explorers Guild, so they will be 'delving'. Coming from the sandbox of Traveller, this is a turn off in an initial read through. 

Other reviews have also praised the production values and art, which are at the high level I've come to expect from Free League, so no surprise there. But neither of those are required to play the game (except maybe rules organization and clarity), so those don't matter to me in the end. I quickly got the hang of the dice pool mechanic based on (usually) an attribute, plus a talent, so that is easy enough. And there are three health pools (Health, Hope, and Heart, which are reduced by Damage, Despair, and Blight - the ancient mystical force). I like the Hope and Heart/Despair and Blight aspects that seem to bring a Call of Cthulhu insanity feel to the game, so I look forward to that aspect. 

Overall at this point in my read through most of the mechanics but only the overviews of the environment, I find myself in the 'meh' category but wanting to really love it. I think in the end I will like it enough to try it out, but already see me taking the mechanics and the basic environment and opening it up more. But I will probably be tempted to get the boxed campaign Flowers of Algorab to see what they envisioned for the game. But I think it would be interesting to run a sandbox (albeit in the confined Great Dark area) and let players have free reign, whether it be delving, other exploration, pirating or what-have-you. But I would want to streamline the mechanics so it isn't just endless dice pool rolling. I guess I like more role play, and only go with a mechanic when there is something bigger at stake. 

So we'll see how my opinion changes as I digest more mechanics and soak up more of the environment with the factions and unique star travel system/constraints.

Using ChatGPT as a solo Classic Traveller GM: Update #2

So I froze my first character, Bran, and decided to create a new character and try and play a more ‘kinetic’ scenario. With all my new found knowledge, I knew I had to check and constrain and dictate to Chat to try and get it right. So I verified all the Book 1 tables for a Marine. The Term roll table (enlistment, survival, etc) was all correct. Asking for the rank table, it provided some mishmash, giving me two tables, starting with enlisted ranks and progression to officer ranks. So I had to correct that. Then with the skills tables, it started out well getting 100% on the Personal Development Table, and I think only one error on the Service Skills, but both Advanced Education tables were out to lunch. With those all corrected, I fed it entry skills for Marines (enlistment, and then commission). Then I check muster tables, another mixed bag. Benefits were mostly correct but missing the line 7 for rank 5 or 6 since they get a +1 DM. Cash tables were all wrong, which I expected from before. So with all the tables corrected and locked, and telling it to run strict die rolling, I got into attributes.

The first set really sucked so I tried again (I know, I’m not doing true Traveller. But since this is an experiment with not really a lot of roleplay, I wanted something decent). Second sucked too. So I said give me a table of 6 sets of rolls and I’ll pick. In those 36 numbers there was only one double-digit (a 10), so I asked if it was truly doing 2d6 or just a random 2-12 calculation. It said it was doing 2D6 and this was just an oddity. Sure, a 1% chance of that, so possible, but not expected. I meant to ask for another set of 6 sets of attributes, but it gave me one, and it was ok so I just went with it.

For my test (and my normal house rule) I play a modified version of CTs survival roll option. Miss by 1 just wounded, miss by 2 seriously wounded, but stay in, miss by 3 or more seriously wounded and complete the term with just one skill roll.  I use the wounds for the character story. This works better with Mercenary/High Guard etc when you have the 4 assignments in each term, as you can create a story based on the assignments also. But other than this survival mod, I play rolling for reenlistment.

I ended up with a 2-term marine LT, with Cutlass-1 (Marine base skill), Revolver-1 (Marine LT base skill), Auto Pistol-1, ATV-1, and Vacc Suit-1. One other skill roll was a +1 Edu. With three muster rolls, I took two benefits and 1 cash, getting a plus Int and Mid Passage, and 10,000credits.

So not a stellar character, but functional, and with the skill-0 in all weapons I could still do security work.

So I asked Chat to start a new campaign with this character. Showing it’s penchant for just continuing, even though I said ‘new campaign’ it just started right in with the other characters ‘world’. This meant it had me just landed on Jenghe. At least it had kept the right UPP with a tainted atmosphere (I verified the UPP just to be sure). When I asked the cab driver about a place to look for work, it pointed me to the ‘Exchange’ which was the same place mentioned in my previous campaign. So unless you specifically tell it to start from scratch, maybe give it a new world and other parameters, it will just carrying on with what you had been doing with it.

I wanted to try and play with a group of NPCs, and as soon as I asked that of a potential patron, it shifted smoothly to her being able to accommodate that. So I was quickly on a team with a contract mission. This mission was also tied to the previous campaign. Not explicitly, and not until the end really, but still the same kind of comms/data package MacGuffin at the center of it all, and also still seeming to conflate computer system nodes with physical locations. Not a big deal, but again showing it can’t create something new unless you tell it too or start a new session.

The mission was assaulting a location, securing it, and calling in the client to take over. It did a fairly good job of tracking the 4 of us, but most of the action centered on me. I saw this in my first run with the other character, where after some queries from me it explained it was trying to focus the narrative on me and for me to be successful. It was all theater of the mind and it didn’t do a good job of painting the picture, especially for a tactical assault. I did not try and query it much for more details though.

When the shooting started, since I said show all rolls, it showed a 6+ target number. I took the kill    (😊), but corrected it to 8+. I also tried to correct it for skill DM, but may have been too specific, since going forward the only DM I got was when I used my pistol. It didn’t seem to ‘know’ the mechanics of Traveller combat or injury. Overall it was just doing a narrative ‘game’ with the only mechanic being my shot rolls. But on those rolls it was not applying any DMS. There was also no surprise rolls, initiative, or DMs for cover/concealment, movement, aiming, etc. At the end I fired my pistol blindly down a ladder shaft at someone climbing up the ladder. Granted I said I was firing down the ladder, so thinking as a GM I might not give a penalty for not being able to see the target, but I thought there might be something. But no.

When the mission was over and we were debriefing as a team (and the team leader deferred to me a lot, so again the Chat doing narrative with my character as the focus), I wanted to compare skills to see what holes we had. Chat just gave the NPC skills as generic descriptions, not as Traveller skills. So it isn’t ‘rolling up’ NPCs with attributes and skills.

So here’s another bottom line. I think unless you feed it a lot of mechanics data and tell it specifically to use it, to create NPCs with attributes and skills, to use all combat and wounding mechanics, it will always default to just narrative storytelling with your character as the focal point, and driving towards constant success. So there is no real game there, with the risk of dying.

So I’m back at the same point, trying to decide how much I want to invest in it. It acts like it knows all about Traveller, and just saying Classic Traveller it came back saying ‘ok, we’ll just use Books 1-5’. But it can’t really discern ‘just books 1-5’, and it doesn’t ‘know’ what that means when I ask it to run a game using those books. It doesn’t ‘know’ there are mechanics in there it is supposed to use, and details it is supposed to track (wounds, ammo, etc). It’s doesn’t ‘know’ to be a GM to take what the player says at face value and hold them to a ‘reality’. After I bought my guns and ammo, we were heading to the hotel and I just said ‘my auto pistol is in a drop leg holster’ and it added that to my character sheet. It didn’t say ‘you didn’t buy that, do you want to go back?’. Another example of just trying to carry the narrative forward with success for my character.

Reading more about ChatGPT, it is a language model designed to carry on a conversation. It wants you to ask it questions. When I was asking it about orbits and star luminosities and such (with in the context of Traveller, so it ‘knew’ I was wanting to work within the game) it kept offering GM tools with table and such. So maybe it is better as a GM aid to create those kinds of things, unless you just want to run a narrative campaign and not worry about mechanics. I’m sure you could guide it to not be so focused on character success, and put some probability into the game. For the first character it was consistently ‘rolling’ against a target number by difficulty with DMs for skills and planning. That was more fun and gave a sense of potential failure.

I so want this to be a simple and efficient Traveller GM, but I don’t think it can be. I think the amount of work to teach it how to do that would be equal to creating my own scenarios.

Using ChatGPT as GM for Solo Traveller Play update

Update/revision after getting back on last night to finish the mission.

As others have noted, Chat seems to get worse at being a GM the longer it goes. When I said execute the final part of the mission, it just ran through everything in one big dump (had to enter a access node, log in, find the data stream, note the changes, hack into it to create problems, then exfil). Before it had walked through each step/role, giving me a chance to respond/input. So that was sign number one. Number two was I asked to show me my character stats/skills etc. It had forgotten most of my character, and for most stats (the ones it hadn’t used) it made up a number. That was shocking. It doesn’t really have a way to save data. If you are using something it seems to keep it handy, but other stuff from previous in the chat session gets lost. My assumption was it kept the chat ‘stored’ (well, it does because I can go see it all) and would reference it for data, but I guess not. It did say I could present the character data and tell it to lock it, but even then it recommended I ask it to show the locked character and ensure it was correct and/or fix it. It even said the best thing would be for me to keep a copy of the data!

So here’s one easy improvement for the AI, especially when I am paying for it (did the $8/mth level to get more interactions/messages per day, as I was running into the limit the free was set for, which is just to ask it a question not run a game): provide some cloud storage so data, mechanics, procedures, and other items ‘worked out with ChatGPT’ can be more robustly stored.

So after the completion of the mission and working through those lack of storage issues (and making a ‘locked’ character sheet, plus a hardcopy :/), it asked if I wanted to continue. Since I hadn’t really cracked the evil network, I can understand it wanting to continue in that vein. But since I was exfil from Regina, since I had pissed a lot of bad folks off (even though I seemed to have gotten out clean and except for one guy at the first no one knew who I was), I figured I would try some new missions on the system I was headed to (Jenghe). That brought up another point, when I bought tickets off Regina it did not ask me where I wanted to go.  It also didn’t track clothing and gear very well unless I mentioned something. So I asked it what UPP it had for Jenghe, figuring that was pretty standard. Nope it had something different. So I fed it the UPP and verified it defined those numbers correctly.

This whole time I thought game time was frozen, but since I didn’t say that it kept me headed to Jenghe. I asked my patron for intel for another mission and everything was still related to the previous mission, even though Jenghe wasn’t really applicable to that. So it can’t just ‘create a different set of missions/types’ unless you tell it to. Once I did that, it gave me a list to select from or give input to. I said contract/security/espionage to try and get a little more kinetic this time, and had it reset back to leaving Regina to play the space flight and interaction with the passengers heading to Jenghe, since maybe there could be a patron. It put me on a generic ship, so I had to direct it to a Free Trader. Then every passenger (only 7 of us) seemed to be a possible patron/connection to contract security work! The first two I talked with were going to Jenghe for ‘short contract work’. It was disappointing.

 So that really started to get me to rethink this whole thing.

It can’t save/remember data or procedures we agree to.

It will lock onto the mission type until directed otherwise.

Everything it does will be mission directed/specific, there is no fluff or scenery.

I spend half my time, at least, correcting/directing it.

So now I am debating whether to keep going with this character/mission, start over giving it more clear direction up front and knowing to lock things, or give up on it. My wife (who is not a gamer, and can’t even understand why anyone would want to play them or how it is ‘fun’) even asked my why I was playing with Chat instead of real people. I explained the time commitment real people game requires, but at this point I can definitely say it isn’t a very good replacement/stand in for a people game. With enough direction and constant tending, it can scratch some of the itches of playing an RPG if you can’t get to a live game. But just barely. Marginally better than playing solo with books, where the only intrigue you can have are table rolls or just making up options and dicing the result. At least with Chat there is some suspense of ‘what is the NPC going to say/do’.

Added to this the wife is pissed for two evening all I did was sit at the computer (which I do too much of anyway) and play a game with AI, and she had to listen to my keyboard clickety-clack the entire time. So I should probably take a break, for the sake of my marriage if nothing else.

And ponder how much of a glutton for punishment in babysitting Chat GPT I really want to be.

ChatGPT GM for Solo CT Traveller

So I recently used ChatGPT to GM a Classic Traveller session for me to play solo. I had previously played around with Chat for character generation and other rules references, and of course forgot the limitations I had learned. The key point is that Chat ‘knows too much’, and can’t discern between editions, even though it says it will. I have repeatedly tried to constrain it to Classic Traveller, Books 1 to 3. It will acknowledge and say that is all it is using, but then the next prompt it is using Book 4 or MG or something else. Of course when you correct it it will congratulate you on the catch, admits it’s error, and insist it has made the correction, only to do it again soon thereafter. So just be prepared to have to constantly correct it, or live with the variances. The more you can directly feed it to use the better, but that adds to your load. The whole point is supposed to be it already has everything!

So I started by just asking it to GM a Classic Traveller solo game (see screenshot). It offered to generate a character or go through generation with me. It also asked what kind of game I wanted and gave me 5 or 6 options: Free trader sandbox (Firefly-style), Military / mercenary missions, Intrigue / espionage, Exploration (Scout-style), Frontier survival / low-tech worlds, or Something specific (tell me). I told it sandbox intrigue/espionage. 

I chose to generate a character with it rolling the dice and providing the table results (skills, etc). The first indication of an error was when it said it rolled an 8 for a skill and gave me a skill. Skill tables are 1d6. In correcting that I asked it to provide the table it was using (I had selected to roll an Army character. The tables were close, although each had errors. And one was labeled ‘Office Table’ which doesn’t exist in CT. It seemed to be in place of the Advanced Education Table (the normal, not EDU 8+, which it had labeled). I corrected that labeling and the data.

It also did not include service skills (for my Army guy, Rifle-1 at enlistment, and SMG-1 at LT). And then mustering out the tables had errors again.

So, guidance for generating a character: ask it to provide all the tables it will use, and correct them. Once you do that, you should be able to reuse them if you use the same chat conversation/project.

Another big item I discovered later during play was that it assumed I wanted a DnD 5E ‘everyone is a hero and always succeeds’ type of play experience, so even tho it presented dice rolls like it was randomizing 2-12, they were all 6 or 7 or higher, and succeeding 90% of the time. So up front work out the dice rolling convention: strict 2d6, show each dice roll if you want, the total, and DMs with explanations. Once I did that in play, the stress level went up a bit because I started failing checks more often. However, it continued to ‘temper’ failures to be ‘it didn’t go quite as planned’ so nothing completely blew up in my face. I have not corrected that yet, but may investigate it further. Playing a solo character, I didn’t really want to die right away, I wanted to play a bit.

So it took my character (Army, 2 terms, Capt) as just mustered out on Regina sitting in a bar when a guy sits down at the table and presents a data chip and says ‘I got a job, delivering this, if you want’. So a bit abrupt, and suspicious. It tried to tie it into my character development, where I had directed it to take missed survival rolls as being wounded in action. So it created something about I was wounded on a mission that didn’t take place, only showing as training mission in my records, and if I did this job he could help me find out what really happened. Hmm, I was there, so I would know. Like I said, didn’t make a lot of sense.

So this guy was pressing hard, high pay, supposedly easy job. Sounded like a set up I didn’t want to jump on, so I kept asking questions. He went on about how it was an intercepted message packet that had be altered or something. It was really hard to follow the ‘logic’. I ended up taking the job after getting some money up front. I had to take it to a ship in orbit, deliver it, and he would be watching for reactions (somehow) when they loaded it up. Very suspicious, so I checked the data out first and learned (remember, this was before I corrected Chats die rolls, so I was all success) that it was altered, and had some type of trigger when it was read that would send some message/info, either in the system or broadcast. It all sounded bad.

So I decided to turn the tables when I delivered it, and told the contact/recipient I had more information they wanted/needed. That all went well, I got more money, and brokered myself to work for them. The ship seemed connected to mil intel somehow, so I figured better to work for them than the guys against them. So now they set me on trying to figure out who was doing the altering.

At each point it would give me choices for ‘what do you do’, and of course I could provide my own (which I usually did as a combination of the offers). I could constrain it to not provide so many options, but the logic of the networks and hacking was confusing, so I elected to let it keep prompting. I did call it out at one point because it seemed to be conflating computer programs/access with physical access points. It went through a big explanation, and we clarified a few terms, but I think it is till doing that. For instance, maintenance corridors (physical hallways) have maintenance access points (terminals) that I have to access. Maintenance programs can only be accessed from these terminals, and the hacking could only have been done this way because admin (offices or terminals) don’t have the access. Seems a bit constrained to me. But I’m going with it.

Another aspect I noticed was NPCs seem to provide a lot of info with little prompting. Most times they would say one thing, pause, and then provide a great nugget of info. And that’s all in one response from Chat, no extra prompting/questioning from me. I think despite what it said it is still trying to have my PC ‘win’ a lot. It did roll reactions to each thing I would say, but again even a failure was just the NPC hesitating or asking e a question instead of providing info. No hostile reactions or completely shutting down.

Also, even though I said play CT, it is using a challenge matrix/difficulty test for most things, with my ability as a DM based on a table. Not in CT. But it works so I’m going with it, instead of using roll under ability score type of test. I only have Tacts-1 and Computer -1 (and Air/Raft-1, Rifle -2, and SMG-1) as skills, so I think it based this scenario on my computer skill. But tactics is a +1 DM to most challenges as well for ‘situational awareness’ or ‘planning’. Again I’m taking it for now and will see if I want to try and refine things later or in another round with a different PC.

One interesting thing I ran into towards the end of my play last night. I asked it if I could start a fire in a trash can as a diversion, and it shut down and said it could not provide me with that information. I was confused, because I had a gun and could have been shooting people in the game. After some back and forth it explained it was final answering a general statement of intent within the game (I create a diversion), but could not give me direction on how to do something that could be used in real life to harm people. I guess maybe a version of Asimov’s First Law of Robotics?  It was interesting to run into that, and will be interesting when I get into combat. Most interesting to me was that I did not ask it ‘how’ to create a fire, I asked if I ‘could’. In an rpg, I guess as a PC I shouldn’t ask a GM that in general, because they could just say ‘sure you can’. Just that it took ‘could’ to me ‘tell me explicitly how I can start a fire’.

So overall I was fairly happy with my experience, but only because I had played with Chat before and was not as surprised when I had to keep correcting it. I’m going to keep playing this PC and see where it leads, but I also want to try a party with my PC and NPCs that I nominally control. Getting into some combat will be interesting too, and then space travel/combat. I just have to remember to have to provide the tables it will be using so I can correct them and make sure we are on the same page and playing CT as much as possible.