Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Radsoc]

Performance wise you would only need a super computer when training the nets. When that's done, it's probably much more light weight (CPU wise) than any mechanical AI. From that point on each game can be used to train the AI on the go. Poorly defined problems is what make stochastical algos and neural nets shine as compared to standard (tailored or not) algos. The problem is that it's a different branch of science with almost no everyday commercial use. A company is not going to pay a programmer for implementing something in 10x the time with everything but guaranteed results where standard algos perform acceptably. But my point is that we've reached the limits of standard strategy AI and that the toolbox of complexity science is the way forward for AI.

[Loud Whispers]

Stellaris does this too, to a point, doesn't it?  I've never really noticed much difference though, since I think most of the difference manifests as opinion modifiers, but a few personalities declare war more readily.

It says they do this, but I've never noticed. I think this is because the power discrepancy between Empires is either too close or too far to make a difference

[E. Albright]

The problem is that it's a different branch of science with almost no everyday commercial use.

This is flat-out wrong. Statistical AI is currently and increasingly applicable to everyday commercial use. Machine learning is everywhere. Big data analysis is everywhere. 10 years ago the everyday part might not have been true, but even then the commercial part definitely was not.

It's not "database programming" common, but we're not talking about some obscure field practiced only in the rarefied atmospheres of academia.

[Radsoc]

The problem is that it's a different branch of science with almost no everyday commercial use.

This is flat-out wrong. Statistical AI is currently and increasingly applicable to everyday commercial use. Machine learning is everywhere. Big data analysis is everywhere. 10 years ago the everyday part might not have been true, but even then the commercial part definitely was not.

It's not "database programming" common, but we're not talking about some obscure field practiced only in the rarefied atmospheres of academia.

That's what I mean, most CS, or other, graduates don't deal with AI. Only a subset works for businesses that do AI, and less are directly involved. I mean e.g. Amazon is a big company, but the people who write recommendation algorithms etc are a tiny part. Most employers want the "database"-part of it, and that's where the main competition is. Then you have the high risk startups and big corps that have their own research departments.

[E. Albright]

That's true of ANY AI programming though. In fact, given how statistical analysis methods have proliferated through various hard- and soft-science research fields, statistical AI techniques are far more accessible than rationalist techniques ever were - and there are ever more tools making them even more accessible. But that doesn't change that AI gameplay which requires planning with deferred results in the face of unpredictable and random changes is a bad fit for these techniques because of how large the feature space is and more importantly how messy the solution definitions are.

Ipsil brought up the biggest impracticality of this approach, though, and it's one that you'd face even if you could frontload all the computing so at runtime it was just applying decision trees: all that upfront computing would need to be re-applied after every patch. Go has a simple feature selection space and limited available actions, so even though the solution space mushrooms massively, it could be tackled with such an approach. But that's because the rules are static. If Go was getting significant rule changes every few months within a non-trivial feature set and to a broad range of possible actions, it would very emphatically not be an attractive candidate. And that's where 4x games sit.

[Radsoc]

This is closer to Stellaris than Go:
https://www.youtube.com/watch?v=tfb6aEUMC04
And very recent. Stellaris is a bit more complex, but here you have unpredictability, planning, unit placement and randomness introduced by the player(s). It still is extremely robust and beats standard AI (which in essence is made obsolete in this game).

Patches and game mechanic changes would make it impractical, yes. However, it doesn't necessarily have to mean that the net wouldn't be robust enough to handle it. In a worst case scenario you would have to retrain it overnight or more. Even with patch changes it is the only attractive candidate if someone wants decent AI.

So, basically, if you developed AI for that game, would you continue to improve it, knowing that efforts are bringing diminishing returns, and changes with patches as well, or go with OpenAI? What if it was possible for Stellaris/Clausewitz? I don't think it will be done, but it's the principle. If the strategy game companies of today are to survive, they have to take a risk.

[E. Albright]

In a worst case scenario you would have to retrain it overnight or more. Even with patch changes it is the only attractive candidate if someone wants decent AI.

"or more" is doing heavy lifting here. The techniques you're discussing need good human players to train off of - OpenAI is certainly relying on the availability of them. So every time the game significantly changes (which per the Paradox model is every DLC), you'd need to get a stable of very good players with various playstyles to generate training data and for the nascent AIs to train against. That would likely be weeks if not months. Every time you make major changes. This just doesn't seem viable or fiscally attractive in terms of ROI - the industry push is "good enough" rationalist AI and let MP take care of the people who want a human-level opponent. The time and resource requirements for machine learning-based AIs seem like they'd need to drop significantly before they became attractive to devs.

[EnigmaticHat]

Yeah but that AI isn't sufficient as an enemy AI in a video game:

1.  It is actually looking at the screen the same way a player does
2.  It doesn't act convincing (see: places sentry wards under towers)
3.  Its way too good/ruthless (because its been trained to the standard of winning, which isn't necessarily a good goal for a video game AI)
4.  DOTA 2 had to be altered for the AI to be able to play it, in particular both teams were given 5 invincible couriers, because the AI could never learn how to use them.  Additionally the hero pool was limited to nukey heroes that didn't scale well off agility*.
5.  Stellaris is far less friendly towards this kind of AI.  DOTA 2 has a static map, static scenario, and much like chess the game state consists of a series of values that can be measured on a good/bad scale.  This isn't true in Stellaris because in Stellaris most assets a player can have are both good and bad, because the game state involves variables that can vary in number instead of just value, and because the game state cannot be interpreted without understanding the intents of the players involved**.

*both the courier and "hard carry" heroes are inherently hard for the AI to handle because they require it to make value judgments between which of its 5 heroes are most valuable.  The courier would be especially difficult because it provides only a positional advantage (something the AI would be weighted not to care too much about) but risks feeding a massive bounty to the enemy team (something a DOTA 2 AI would definitely be weighted to hate).

**hence why Stellaris is somewhat like a board game.  Its the same reason "OpenAI Catan" would be an exercise in madness.

[Radsoc]

I think it trains against itself, and then goes a match against real players. But they say the paper isn't out yet.


This one doesn't seem to look at the screen the same way a player does. (No need for image recognition - and then you would probably have to strip graphics down a lot). I think it has direct access to the game. If it acts convincing or not is difficult to say and hard to measure. In the video they mention that it seemed to act suboptimally/make mistakes (but which eventually made sense). It will probably exploit mechanics if there's an opportunity, but that's the fault of the mechanics and the incentives supplied there really. If it's too good, you just grab it at a smaller generation number. They mentioned that in the vid. Early on it would only beat casual players. The thing is that you can handicap it, while you today have to handicap yourself. Victory goals and personalities could be accounted for in the same way, by adjusting goals.

Stellaris is a slower game though, but if time is measured as interactions/events per minute Dota could be in the lead. The strength of nets is robustness to peculiarities of games. I believe this is doable in principle, but yes, I agree there are too many practical problems, but I believe that if Paradox took a small chunk of the money they get, they would get their investment back in a decade.

I stopped playing lots of strategy games in SP due to AI. In civ 5, in 2/3 campaigns I never lost a unit (except cruise missiles). In Shogun 2 I lost 2 battles out of 100, and they were auto. Dominions 4/5 is almost unplayable in SP, apart form a trainer for MP. If Stellaris didn't have story driven events or mods I would only play it for the challenge of learning the game and then leave it. Right now the Star Trek mod is excellent.

[E. Albright]

They said they trained it both against itself and players. It experienced a serious bump-up in quality when it got to train against players.

I don't know enough about MOBA games to make any comments on the complexity involved, but what EnigmaticHat says points towards this being a pretty bad parallel even if we ignore the rules being crippled to handicap the players' advantages. There may be a lot of moving parts, but not as much complexity in how those interact with each other or the environment - and the specific sort of necessary handicaps (i.e., involving non-straightforward play) points towards this not being good evidence of parallels to a PITA complexity like Stellaris.

[Telgin]

New dev diary focuses on performance: Dev Diary.

Of note, Stellaris will be 64-bit (only) in 2.3, and performance is expected to be 10-30% better.  The devs also took some AI weighting from Glavius's mod to improve the AI.

[Telgin]

Patch notes for 2.3: link.  Next patch + DLC will be released on June 4th.

Copied for quick access:

Spoiler: Patch Notes (click to show/hide)

#################################################################
######################### VERSION 2.3.0 ###########################
#################################################################


###################
# Ancient Relics Story Pack Features
###################

# Relics

* Collect Relics to show off the achievements of your empire. Display your glory by showing off your very own Omnicodex, Miniature Galaxy, and more. All grant powerful passive and active effects to their owner.

# Archaeology

* A system that allows multi-stage investigations of strange alien sites, with rewards for unlocking each stage. Find them, excavate them, make their secrets yours. Also added some new archaeology-themed scientist traits tying in to these.

# Two New Precursors

* Learn more about the ancient history of the Stellaris galaxy with two more Precursor races: the Baol, a peaceful plantoid hive mind, and the Zroni, a powerful psychic species that was the first to reach the mysterious alternate dimension known as The Shroud. Find their ancient homeworlds and hopefully learn from their demise.

# Relic Worlds

* Discover a new world type, a mausoleum on a planetary scale. The the dead civilizations that once dwelt here left behind a treasure trove of arcane technology, which you can excavate at archaeological sites.

# Minor Artifacts

* Ancient Relics adds a new rare resource, Minor Artifacts, which can be uncovered via archaeology and exploration. They can be spent to give powerful empire ethics-specific boons. Celebrate the diversity of life in the galaxy by displaying these artifacts in a museum, or destroy them to proclaim the superiority of your own eternal empire.

# Achievements

* Unlock the mysteries of the galaxy to get 10 new Steam achievements


######################
# 2.3.0 "Wolfe" Free Features
######################

# Prior Content Relic Rework

* Many things from previous game content have been reworked to use the new Relics system. Among several others, you can gain the Isolated Contingency Core or the Prethoryn Brood Queen as symbols of your triumph.

# Sector Redesign and Automation

* Sectors have been greatly improved in 2.3, allowing the player more control over how and where they are formed, as well as giving more robust automation tools to let them run themselves if you want to focus on the bigger picture.

# Planet Designation

* You can now choose to manually set a planet designation (Agri-World, Forge World, etc), focusing production on different resource outputs. There is now a shared automated development stockpile that you can pay in to and allow planets to set their own build orders.

# Megastructure Improvements

* Habitats gain different effects based on the type of celestial body they inhabit. Ring Worlds gain specialized, highly efficient districts.

# Other Stuff

* 2.3 is now a 64 Bit application. Support for 32 Bits is deprecated, but you can roll back to 2.2 if you want to play on such hardware.
* Precursors no longer spawn in a preselected pie slice of the galaxy
* Added a menu option at game start to disable Caravaneer fleets if you don't want those sweet, sweet deals
* Added a randomize button when renaming planets
* Rogue Servitors can now build a variant of leisure districts on habitats to entertain their bio-trophies in space

###################
# Balance
###################

* Thrall Worlds can now build Energy Grids and Mineral Purification Plants
* Terraforming worlds no longer re-rolls districts, blockers, unique deposits
* The Voor Technocracy prescripted Empire is now Fanatic Materialist/Authoritarian
* Technocracy civic now requires fanatic materialist ethics
* Rebalanced the Worm-In-Waiting to make it more of a real threat
* Habitats can now build food processing buildings, research labs, commercial zones, and holo theaters
* Adjusted costs of clearing Arcology blockers
* Habitats built over rare deposits now gain the associated deposits
* Mastery of Nature ascension perk requirement reverted, can once again be picked in the first slot
* Ecumenopolis districts now require rare resources to build (50) and upkeep (2)
* Low habitability now also reduces pop growth and job output by 0.5% per missing habitability
* Gaia World terraforming time reduced from 7200 to 3600 days, cost reduced from 10000 to 7500
* Matter Decompressor Site alloy cost increased from 3000 to 5000, to be consistent with other sites
* Architectural Renaissance unity ambition now also increases megastructure build capacity by +1
* Dyson Sphere energy output increased from 250/500/750/1000 to 1000/2000/3000/4000
* Matter Decompressor mineral output increased from 250/500/750/1000 to 500/1000/1500/2000
* Consolidated placement rules for most megastructures, so that they are no longer inconsistent
* Interstellar Assembly now costs energy in upkeep instead of consumer goods, restoration alloy cost reduced from 20000 to 15000
* Mining Drone Home Base now also drops a strange device
* Gestalt races now also benefit from increasing the flow of Dark Matter from the Dimensional Portal
* Rogue Servitors start with another Nexus District
* Galactic Wonders ascension perk no longer immediately unlocks the ability to build Ring World, Matter Decompressor and Dyson Sphere, but instead adds them as research options
* Picking life seeded will no longer grant neighboring Gaia worlds, but rather a randomized starting planet class
* Clerks and Preachers job weights no longer care about being strong, weak, or proles
* Miner job weights DO care about being proles
* Synth Empires pay only alloys for Colony ships, like Machines
* Inward Perfection empires can no longer form research agreements or commercial pacts

###################
# Performance
###################

* Fixed a case where unneeded VFX were being rendered, causing a gradual game slowdown over time
* Reduced the delay when there is a big transfer of systems after a country surrender
* Fixed the end of year lag when big empires with lots of pops being assimilated convert them into proper citizens
* Improved faction threading, for performance and OOS safety
* Smoothed border calculations to reduce stutter
* Reordered threaded economics calc to avoid race condition of ships depending on planet resources depots
* Fixed a case where every pop would check for faction membership on every daily tick, creating a performance drain

###################
# UI
###################

* Corrected misleading tooltip that falsely claimed it was not possible to set a species to Transcendent Awakening when it has Assimilation citizenship
* Queuing up multiple outpost constructions will now give the adjacency discount for influence costs where appropriate
* Fixed an issue when having multiple fleets selected including a colossus, right click orders menu on planets would be empty
* Rivalry influence production now displayed with modifiers in diplomatic view
* Diplomatic actions now display correct influence upkeep cost with modifiers
* Added entry to the fleet capacity tooltip explaining how many Titans you can build
* Fixed terraform button tooltip wrongly suggesting you can terraform gaia worlds
* Fixed some technology description texts overlapping with the tech specialization icon
* Fixed quite unplayable text spillout for guaranteed habitable worlds slider in multiplayer galaxy setup
* Fixed info about "Mass Produced" and "Custom-Made" modifiers for robot assembly being missing in the tooltip
* Fixed the reinforce fleets option in the Fleet Manager wrongly showing missing minerals rather than alloys, if you don't have enough of the latter
* Fixed text spillout of View Privacy Policy button in options (I'm looking at you, 23 letter German compound word)
* Fixed a heinous pixel error with Mass Driver I
* Fixed fractional costs of rare resources in ship designer displaying as 0, causing confusion
* Added tooltip to terraform button when it is disabled, detailing why
* Removed redundant repetitive redundant diplo offer duplication for peace offers
* Fixed an issue where the overall progress of a fleet upgrade order would visually reset to 0% after any single ship in the fleet completed upgrades
* Fixed observer outliner to be less broken, along with the hotjoin interface
* Fixed missing species portrait in the declining pop window, so now you can feel extra sad
* Fixed a display issue where replacing a built district would temporarily add to the number of built districts, sometimes causing overflows
* Replacing a planetary building now triggers a re-sort of building icons to keep similar types grouped together
* Removed the "has no FTL drive" part of the tooltip for stations when they are selected and you hover over a different system
* Fixed an issue where losing a building due to resettlement would show the pop requirements for all slots displaying as 5 lower than reality
* Even though devouring swarms can now purge inorganics for alloys, the tooltip wrongly stated they could not be processed...fixed
* Fixed so that unemployment by strata icons are properly displayed and has a working tooltip
* Now it is possible to add federation ships to a fleet design
* Fixed commercial pact diplo agreement wrongly inverting what the proposing and receiving empire will get from the deal
* Fixed visual issue wrongly showing zero tech progress until after first monthly tick
* Cornucopicturesque and Mineralistic event option tooltips no longer refer to adding tile deposits
* Fixed placeholder icons for Ascendancy Shipwrights modifier and Menial Drone Output (extra % produced by Coordinators)
* Fixed missing reference to influence gain reduction in the tooltip describing the result of losing a humiliation war
* Fixed unplayable overlap of the investigate button on unknown alien fleets
* Fixed bad text wrapping in federation fleet power display when the number was over 10000
* Machine empire Mineral Purification/Energy grid etc buildings no longer show the tooltip for raising district cap on inapplicable planet types
* Fixed tooltip that falsely claimed that setting a species to purge (except neuter) was a requirement for the Decent Living Standard
* Habitability map icon now considers all possible species instead of just owned ones
* Added descriptive text to the jobs screen to explain how favouriting jobs works
* Updated Corporate Dominion civic tooltip to refer to its new post-MegaCorp effect
* increased tooltip decimal precision on the complex drone output bonus Rogue Servitors gain from Biotrophies

###################
# AI
###################

* Implemented improved AI building selection weights (thanks to Glavius for awesome insight here)
* AI will now properly bombard and invade primitive planets rather than suffering last minute pangs of conscience about using orbital lasers on people armed with bronze daggers
* AI now checks for how many outposts are being upgraded to starbase status already when deciding if it can build more
* Fixed the AI considering federation ships as part of its own fleet capacity, causing them to drastically neglect their own fleet building
* Fixed a bug preventing sector automation ever upgrading research labs to level 3
* AI will not build nanite transfusers without nanite income, nor build more than 2 per planet
* Fixed AI fleets getting stuck with redundant attack/defend orders if their target went MIA
* The sector AI of Machine Intelligences and Synthetically Ascended Empires will no longer build food-producing districts or buildings unless they actually need food
* AI will now consider repairing ruined branch office buildings
* Fixed so that transports don't emergency FTL as soon as they take any damage whatsoever, because D-Day didn't stop when the first landing craft hit a mine
* Fixed some edge cases where the AI would attempt to build its way out of unemployment with districts and buildings that its pops can't work
* AI prefers City districts over Housing buildings a bit more
* Machine empires will no longer build nonsentient, less efficient robot pops and unemploy their own citizens
* AI will now take more opportunities to repair damaged fleets when it has the chance
* Fixed AI bug caused by fleets unassigned to any group not being able to be added to one due to damage, instead just sitting idle
* Fixed AI not bothering to send its war-weary fleets home following a peace agreement

###################
# Modding
###################

* The Archaeology system from Ancient Relics is open for modding use, to create your own multi-stage event chains
* Fixed a map distance cache bug that affected mods with non-random maps, causing each outpost to cause the maximum amount of influence
* Added GetAge, GetFirstName and GetSecondName to the leader scope in localization, to allow you to do events calling just first or second names, etc
* on_fleet_destroyed_xxx events will now happen only once. For example: the shattered loop
* Added num_owned_relics trigger
* kill_pop_chance now uses weights with Pop scope
* Added locked_random_list with predetermined effect for random colony automation
* added building/district specific building speed modifier "planet_<building/district>_build_speed_mult"
* Parallel Megastructure build/upgrade capacity can now be altered by modifiers
* Added effect "add_timed_trait" that lets scripts add a timed trait to a leader character. Usage: Scope Leader, `add_timed_trait = { trait = <trait name> days/months/years=<duration>"
* Fixed the "num_buildings" trigger to actually take the paramater "any" into consideration
* Added "show_prereqs" to megastructures to determinie if they appear or not in the tech that unlocks them
* Added "prereq_name" to change what name a megastructure shows when unlocked by a tech
* Added trigger last_activated_relic
* Added has_deposit_category for checking a category of a deposit
* Added has_relic trigger
* Added remove_relic effect
* Added add_relic effect

###################
# Bugfixes
###################

* Fixed a case where Prethoryn Infestors would not infest habitable worlds
* Fixed a case where the L-gate chain could get stuck, with pointless repeating techs to gain insights, if another empire finished the chain first
* Fixed nonfunctional "upgrade all" button for defense stations
* Migration controls will now actually control migration, like it says on the tin
* Fixed a case where choosing to fight the Worm could spawn another one, that could also be farmed for rewards
* Fixed broken starbase designs for Awakened Fallen Empires that prevented them from expanding
* Fixed a host of display and stability issues relating to constructing robots from a template rather than the base type
* The Caravaneers will no longer offer the Bunk Beds and Food Containers deals to empire types that cannot use them (Gestalts, etc)
* Put in safeguards to prevent game events dependent on contact with another normal empire (The Prince, Legible, etc) from firing before you've made first contact with any
* Cybernetic pops should now be able to survive (poorly) on Machine Worlds, which now have a base habitability of 0 (offset somewhat by their +20% tolerance)
* Fixed organics being assimilated not being put in a proper job/stratum leading to no assimilation progress at all
* Fixed a bug when pop growth penalty due to low habitability was not applied
* Fixed certain binary and trinary star systems with planets visually inside a star
* Fixed the bug when state of truce was not force-opening borders
* Fixed unintended case where borders force-opened by a peace deal could have trade routes sent through them, creating weirdness like piracy in gestalt empires, etc
* Fixed ironman cloud saves being overwritten if you play the same empire in separate games
* Fixed the "Old With Defects" event tooltip stating it would grant progress in one technology, instead applying it at random to one the player already had available to research
* Fixed a crash on Mac when attempting to load a game with the resettlement screen open
* Fixed literally unplayable typos
* Null void beams no longer wrongly appear outside of the Portal event
* Fixed synthetic empires not getting the project to stop the Ghost Signal
* Replaced placeholder icons for certain ship/station types in fleet combat results, etc
* Fixed Volatile Mote production jobs for machine empires so they are now properly complex jobs with higher priority for pop assignment
* Added a research cost to the "Cracking it Open" event, also changed it to be an engineering project
* Added a research cost to "Breathe In, Breathe Out"
* Fixed a case where you couldn't build an outpost in a system with hostile primitives because it counted as enemy territory, nor could you invade and subjugate them to rectify this situation because they weren't inside your borders, creating a weird catch 22
* Fixed a confusing visual issue when using manual ship design where going to the ship designer would automatically show newly researched components, even if your ship doesn't actually use them
* Fixed the penalty for going over your starbase cap only being applied once you were 2 starbases over the limit
* mature_galaxy and advanced_galaxy console commands are functional again
* Fixed a crash on game save
* Fixed a free alloys exploit with downgrading defense platforms
* Fixed planetary features with blockers that are cleared by an event not granting districts properly until after a game reload
* You can now choose to station military ships at things like wormholes the same way you can with civilian ships
* Fixing a mysterious Linux crash when colonizing a planet and the species somehow gets invalidated before finished
* Crashfix when invalid leaders are left in a country's leader pool
* Fixed a low stability warning on game start that disappeared after the first daily tick, because the apparatus of control and repression cannot be relaxed for even an instant
* Fixed some cases where special projects could get stuck
* Fixed an uncanny crash with clicking on the ethics icons of the UNE when editing the default empire template because your utopian, egalitarian ideals aren't as universal and immutable as you think, hippies
* Added missing piracy suppression values to Caravaneer-granted ships
* Fixed weird cases where genetically modified subspecies could unintentionally be enslaved with certain ethic/civic combinations
* Fixed unusually slow purge speed when the Prethoryn conquer a primitive world
* Failing to Preemptively strike the Underground civilization as a homicidal race will cause them to invade you rather than just do nothing
* Fixed a crash when giving away systems
* Fixed the trade window sometimes becoming crazy until reload/restart
* Speed Demon event will add the red/green traits even if your species is very strong/adaptive
* Fixed OOS for diplomatic upkeep upon hotjoining
* Fixed a hilarious one where marauders could demand the outrageous tribute sum of zero resources
* Fixed the Odd Factory event creating incorrect worker strata jobs instead of drone jobs for gestalt empires
* Ships will no longer use unexplored wormholes when trying to pathfind
* Fixed a bug where ruined buildings could be downgraded, repairing them, even though it's true that sometimes you have to destroy in order to create
* Fixed Life Seeded not granting proper Gaia world preference
* Changed "The Planet that Was Not There" event wrongly targeting guaranteed habitable worlds, nor on gaia planets or those with special deposits
* Fixed at least one instance of the country resource OOS
* Ecumenopolis preference no longer called AI world preference
* Crime event no longer fires for non-sapient robots, and also waits until the planet has >3 pops
* Fixed a Mac crash when trying to remove a build queue item from a starbase
* Fixed slightly misleading description text if the first gateway related site you discover is the L-Gate
* Randomized empires gets their starting planets, ethics and civics correctly set up when they start as life seeded or post apocalyptic
* Fixed breaking commercial pacts not incurring a negative diplomatic modifier
* Fixed the mining restoration project failure result giving a scientist trait to an admiral. England expects every man to do his duty.
* Fixed Machine and Hive world terraforming removing rare deposit types
* Fixed a wild one where multiple research point consuming special projects could put your research output in to negative, and block all station or building construction for some reason because of "missing resources"
* Fixed missing event image for the Nanosand anomaly when spawned on a desert world
* Fixed a crash when a country gets a trade route that used to belong to someone else
* Fixed planets under colonization wrongly generating unrest warnings
* Impose Ideology war now properly creates an independent allied empire, instead of a vassal
* Fixed terraforming candidates not being terraformable into Gaia worlds

* Lots of general crash, OOS, stability fixes

I really like most of these changes.  Making habitability have more teeth is a good thing, although it means I probably won't try any more Life-Seeded games for a while.  The changes to habitats and ring worlds were also long overdue.

One thing that kind of confuses me is how much of a buff the Dyson sphere got.  I'm not sure what the point of generating so much energy is, but I guess it means I'll have to invest less alloys in them now and just upgrade them one or two stages.

[Cruxador]

Gotta say, I don't much like that the whole system of archeology (rather than just the content that currently uses it) is locked behind DLC. Hopefully they change that after adding more relevant content in future releases, rather than just leaving it as an isolated system and not iterating on it.

Regarding the Dyson sphere, they may have found it not worth investing in on their own office games, but it has been a complaint since the inception of the megastructure system that the Dyson sphere is not a big enough deal. Since this should allow it to actually serve your empire's entire energy needs on a non-small map, this seems about right to me.

[Telgin]

If I understand right, the archaeology system is actually part of the free patch and will even be available to modders for free with 2.3.  I think it's just the DLC content that will really use it at first.

And that may be right about the Dyson sphere, and to a lesser extent the matter decompressor.  I always end up building both and they're never enough on their own.  I found myself still building energy districts in the end game in my last game even with Capacity Overload, and actually got to the point of having no net mineral production because of alloy and consumer goods consumption.

[Broseph Stalin]

I've always used mods for multiple dyson spheres, allowing them to actually sate the energy thirst of a mature empire seems like a good call.