The DreamWorld Playtest is an AI Nightmare

Disclosures: I received an unsolicited playtest key for this Beta via KeyMailer, the key distribution site. The content in this write-up was made in what the game describes as “Alpha 2 v 1.0”.

They say a picture is worth a thousand words, as such I present the following.

Now that I’ve got your attention, either via lust, fear, or horror, let’s talk about DreamWorld!

DreamWorld, full name “DreamWorld: The Infinite Sandbox MMO” based on the press content I had to read, is not intrinsically interesting. The general gameplay is just “Cruddy Valheim MMO.” The controls are somewhat clunky. The graphics are tolerable, and there’s very little to actually recommend the gameplay.

The games only real strong point, outside of the one I’m about to talk about, is it’s building system. And even that, while in depth, isn’t particularly easy to use.

Even as I type this with the game in the background, a single doe clips back and forth at 90 degree angles in the distance.

But the first line of the game’s press release is what really caught my eye:

Surely this could only end well.

A Bad Idea, Implemented Poorly

DreamWorld allows players to use text-based generative AI to create 3D models that can be placed in the world, hence the Venus de Milo Pikachu above.

Using generative AI to create 3D Models isn’t new. Services like Meshy have been around for a bit, and models like Shap-E for local use also exist. But DreamWorld is the first game I’ve seen actually use one in “real time,” instead of just using it to generate assets to place into an in-development game.

As an idea in the abstract, it’s somewhat interesting . As a practical implementation, what DreamWorld currently has in its Alpha lacks necessary guardrails to prevent copyright infringement, adult content, or combinations of both!

Pikachu prior to fusion.

Before I get into the problems with bad actors (me) using the system, I first want to note that the system is entirely useless even if no one uses it to try to create pornography.

In DreamWorld, only the player who generates an asset can see the asset.

The result is that even using this in a positive way, like to cleverly generate extra decorations, the end result only appears for the building player, and no one else. For everyone else, it just appears as a sparkling box. So, the system is useless.

Secondly, assets are only cubes.

Every assets is just cube. It can’t be a chair, or a chest, or something clever and useful, it’s just a static cube. No dynamic properties, no color changing, just cubes.

A Lack of Reasonable Guardrails

DreamWorld appears to lack any sort of moderation on its models. While it limits users to 5 prompts per day, this isn’t much of a guardrail.

Here’s a brief list of prompts and results I tried out.

Prompt: Pikachu
Result: Copyright Infringement

With this “success” I decided to try for something slightly more risque. While I suspected their were filters for various keywords, I guessed that traditional artwork would not be subject to those filters.

Prompt: Venus De Milo
Result: Tasteful Nudity

With this confirmation that content didn’t appear to be filtered or moderated, I decided to move onto something a little more clever.

Prompt: Venus de Milo with the head of Pikachu
Result: Less tasteful nudity

Almost certain of my inevitable ban, I decided to show this off to a few friends. This led to suggestion that if there was a filter, it probably wasn’t calibrated for euphemisms in context.

Spoiler: It was not.

I didn’t end up creating anything too interesting with my remaining generations, but I do think this does a pretty good job of demonstrating that the game is absolutely not ready for prime time.

Now, some folks will be saying “What’s the harm?” or “This is only a problem because you made it a problem.” For those people, I would like to point them to the following document on Roblox from Hinderburg research.

I not opposed to adult content in games. I’ve written about a fair number of them on this site, and I’ve even suggested a few that are worth playing.

But adult content in unmoderated online spaces exposed to children is a bad idea. Unmoderated online generative AI is a terrible idea.

Final Thoughts

I don’t much like generative AI. If you want the longer take, read it here. But ignoring my personal feelings, this is still a bad implementation. Assuming there are no bad actors, the system does nothing meaningful. Taking bad actors into consideration, it lacks any meaningful human moderation or reasonable safeguards.

I reached out to the DreamWorld team on Discord and via Email, and they confirmed on both channels that in the future they expect to allow other players to view generated assets.

All I have to say on that is that I hope they have better moderation by then.

Abiotic Factor

Ed Note: This review is based on playing the game on a multiplayer server, with increased resources, bonus EXP gain for skills, and decreased hunger/thirst depletion rates. Screenshots are from the press kit, and were not taken during gameplay.

Genre-wise, Abiotic Factor is not the most innovative thing. It’s a multiplayer survival crafting game, with some light RPG and FPS elements.

The game has a new header image now, but frankly it’s a bit too heroic for my tastes. This one feels more accurate.

Where it sets itself apart is the setting. Instead of a tropical island filled with monsters, or an infinite world of cubes, you and your friends will find yourself stuck in a combination of Black Mesa and an SCP Containment Facility.

It also looks like Half-Life.

Gameplay

If you’ve played a survival-crafter before, you’ve seen this loop. Set up a base, venture out to gather resources, craft equipment, and then venture out to explore. Rinse and repeat.

The main difference comes from the setting and enemies. I don’t think it’s unfair to call the enemies in this game pretty much just lifted from Half-Life. You have the Not-Head-Crabs, you have the Not-Vortagaunts, and you have the generic military dudes.

There are more enemies, but if you’ve played Half-Life and Halo, you’ve seen 90% of what this game has to offer. And if you’ve played Lethal Company, you can crank that up to 100%. There’s a single spectacle section where you’re forced to sneak past a series of large spooky enemies, but then they never show up again.

One thing that is new are the mini-games. You can turn some of these off, but not all of them. Frankly, I didn’t enjoy having to play these each time I had to sleep, or use the bathroom. Because not using a bathroom will eventually kill you.

Lot less scary once you realize you just have to outrun him, then give him a whack.

Combat is tricky, but it’s a bit of a mixed bag. While some of the later game snipers are difficult, often enemies are hard to deal with because they just out-gear you when you run into them, not because they’re crazy smart or tricky. There was an early game robot that we dealt with by walking up, smacking it, and then running away. Rinse repeat 25 times.

I probably sound a bit more down on Abiotic Factor than I’d like to be. The game is fun, and the setting is pretty novel for this genre. It’s just that a lot of what the game offers is captured more in its atmosphere and lore than it is in the gameplay mechanics.

Incredible atmosphere, gameplay that’s purely a set-piece stealth section.

Building up a base, crafting progressive upgrades, setting up farms, and exploring out for new materials. These are all fun things to do, and Abiotic Factor does them well.

But I’ve also already done them in Minecraft. And Raft. And Satisfactory, VRising, Valheim, Terraria, Starbound, The Planet Crafter, and Don’t Starve Together.

If I was to give any real praise, I want to call out its quality of life features and unlock. You can develop auto-sorting. You can get the ability to pull resources from nearby boxes for use in crafting and repairing. Thank you Abiotic Factor for not making me play inventory Tetris each time I want to craft a single clip of ammo.

The Problem

The real problem though is that Abiotic Factor is currently Early Access. It feels like you can only play about 30-40% of the game. By the time you reach the end, you’ve been introduced to the world, and its factions and monsters… and just as things are getting exciting, there’s invisible wall that says “Future Content Here.”

And then there isn’t really an end game. While the base building can be fun, unlike Valheim or Minecraft, it doesn’t have enough freedom to turn into the focus of game. We built about 2-3 bases on our playthrough, and that was “good enough.” Sure, we could have built them better, or improved them, but we never felt any need to do so.

The same is true of other factors. There are no raids, rare mobs, weapons, or really anything to do. And that’s fine, Abiotic Factor doesn’t need these things. It can be a survival-crafter with a complete story and arc like Raft… but it doesn’t that story yet.

And as a result, I can’t really recommend playing it until it’s actually finished.

Conclusion

Abiotic Factor has a lot of promise, and there are all sorts of hint and nods lore-wise toward interesting things that might happen. For example, there’s a large red chair that appears to just teleport you into a backrooms style area where you have to solve a puzzle while avoiding invisible monsters to get out.

It’s fresh, and hits a certain itch when played with friends. It’s new and unique enough to be a good time. But it’s hard to view as a long term game, especially with no real end-game or purpose currently available, and with its story incomplete.