
Hostile Neural Networks in All the Mods 10: To the Sky
All the Mods 10: To the Sky 2.1.0 ships version 6.5.1 · First seen in 1.0
Hostile Neural Networks is a mod in All the Mods 10: To the Sky 2.1.0, where it is version 6.5.1. Simulate Mobs to create not-so-simulated drops! (Summary from the mod's Modrinth page.)
Hostile Neural Networks guide for All the Mods 10: To the Sky
Last verified for All the Mods 10: To the Sky v2.1.0 on Oct 16, 2026 · text last changed Oct 16, 2026
- Sections
- 6
- Recipes shown
- 10
- Items named
- 13
- Quests covered
- 10
Every recipe and number here was checked against All the Mods 10: To the Sky 2.1.0, read from the running game on Oct 15, 2026.
What is Hostile Neural Networks?
2 parts
Hostile Neural Networks lets you farm mob loot without hunting for it. You build a Data Model of a mob, train it, run simulations, and turn the resulting predictions into that mob's drops. It's based on Deep Mob Learning, so if you've used that, the loop will feel familiar. It's a great fit if you'd rather automate drops than grind spawners.[1]
This pack ships Hostile Neural Networks 6.5.1 for Minecraft 1.21.1.[2]
The basic loop
- Make a Model Framework and right click the mob you want to study. It turns into that mob's Data Model.[1][3]
- Put the model in a Deep Learner and kill that mob while you carry it. One Learner holds 4 Data Models, and it counts kills as long as it's anywhere in your inventory.[3]
- Models start at Faulty and must reach at least Basic before you can simulate them.[1][3]
- Put the model in a Simulation Chamber with power and a Prediction Matrix to run simulations. These always give a generalized prediction and sometimes a mob prediction.[1][3]
- Feed mob predictions into a Loot Fabricator and pick the drop you want. It remembers your choice for each model type, so it can be automated.[1][3]
The generalized prediction a simulation gives you varies with the origin of the mob in the model. The pack's quest book has a quest each for Generalized Overworld Prediction, Generalized Nether Prediction and Generalized Ender Prediction, which it describes as making various Overworld, Nether and End materials.[3]
Where it fits in the pack
The pack has a Hostile Neural Networks quest chapter, and the ATM Star path includes several of its quests. "Learning Mobs" quest and "Modeling Mobs" quest come first. "Simulating Deaths" quest unlocks after both, and "Creating Loot" quest follows that one. The chapter's own examples of loot you can fabricate include Dragon Eggs from Ender Dragons, Iron from Zombies and Nether Stars from Withers.[3]
How do you get started with Hostile Neural Networks?
6 parts
Hostile Neural Networks lets you train a Data Model on a mob, run simulations with it, and turn the resulting predictions into that mob's drops. The steps below follow the quest chapter, so you can build things as the quests unlock.[1][3]
Step one: make a Deep Learner
Start with the Deep Learner. Right-click it to open a HUD where you store Data Models. Each one holds 4 Data Models, and it collects kill data as long as it's anywhere in your inventory. If you carry several, kills count for the models in all of them.[3]
Crafting Shaped
| any obsidian | any obsidian | |
| any glass panes | ||
| any obsidian | any redstone dust | any obsidian |
Recipe id: hostilenetworks:deep_learner
Step two: turn a Model Framework into a Data Model
Next, craft a Model Framework. Right-click the mob you want to study and the framework turns into that mob's Data Model. Put the new model in your Deep Learner, then kill that specific mob with the Learner in your inventory.[1][3]
Crafting Shaped
| any redstone dust | any redstone dust | |
| any gold ingot |
Recipe id: hostilenetworks:framework
Step three: level the model up to Basic
Every Data Model starts at the Faulty tier, and you need at least Basic before you can simulate it. The Deep Learner HUD shows how many kills are left for the next level, and the model's tooltip shows the same info.[1][3]
Step four: build a Prediction Matrix and a Simulation Chamber
The Simulation Chamber is the main machine of the mod. It runs simulations on the Data Model inside, and it needs at least one Prediction Matrix to do so. The power it needs depends on the model, and you can read it in the model's tooltip.[3]
Crafting Shaped
| any iron ingot | any glass panes | Empty slot |
| any glass panes | any glass panes | |
| Empty slot | any glass panes | any gold ingot |
Recipe id: hostilenetworks:prediction_matrix
Crafting Shaped
| Empty slot | any glass panes | Empty slot |
| any obsidian | ||
| any lapi gems | any lapi gems |
Recipe id: hostilenetworks:sim_chamber
There are two simulation modes. Training makes the model gain data but produces no predictions, and it costs more Forge Energy than Inference. Inference outputs its result onto the Matrix, but the model gains no data. Higher tier models have a higher success rate for getting a Mob Prediction.[3][4]
Step five: turn predictions into loot
Every simulation produces some kind of Generalized Prediction, which is used in other crafting recipes. A successful one also gives you a Mob Prediction. Feed that into a Loot Fabricator with power and pick which item you want from that mob's model. The machine remembers your choice, so you can automate it. The quest gives Zombies for Iron and Withers for Nether Stars as examples.[1][3]
Crafting Shaped
| Empty slot | any netherite ingot | Empty slot |
| any diamond gem | any obsidian | any diamond gem |
| any gold ingot | any gold ingot |
Recipe id: hostilenetworks:loot_fabricator
Later: the Data Center
Once the basics work, the Data Center combines the Simulation Chamber and Loot Fabricator in a 7x7x7 multiblock. It handles 25 Self Aware Data Models at once and produces loot directly instead of Mob Predictions.[3][4]
Crafting Shaped
| any obsidian | ||
| any obsidian | any obsidian | any obsidian |
Recipe id: hostilenetworks:data_center
The floor must be obsidian, and the walls and ceiling must be black stained glass. The controller replaces one bottom-wall block. Data Center IO Port blocks can replace wall or ceiling blocks. The pack changed the IO Port recipe, so use this one rather than a recipe from the base mod.[3][4][5]
Crafting Shaped
| any iron ingot | any iron ingot | |
| any redstone storage block | ||
| any iron ingot | any iron ingot |
Recipe id: hostilenetworks:data_center_io_port
How does Hostile Neural Networks work?
6 parts
Hostile Neural Networks is built around one idea: you train a Data Model of a mob, run simulations with it, and turn the results into real mob drops. Here is how each step works in this pack.[1]
Building and training a Data Model
The Data Model is the core unit of the mod. It holds information about an entity, and you use it to generate that entity's drops. To get one, craft a Model Framework and use it on the target entity to build the model.[1][4]
Crafting Shaped
| any redstone dust | any redstone dust | |
| any gold ingot |
Recipe id: hostilenetworks:framework
Every model starts out Faulty. You train it up to Basic with a Deep Learner. It stores a few Data Models, and the kills you make begin to train them. The other training route is the Simulation Chamber.[1]
Crafting Shaped
| any obsidian | any obsidian | |
| any glass panes | ||
| any obsidian | any redstone dust | any obsidian |
Recipe id: hostilenetworks:deep_learner
Running simulations
Once a model is Basic or higher, the Simulation Chamber can run simulations of it. Every simulation produces a generalized prediction. It may also produce a Mob Prediction, depending on the model's accuracy. A higher model tier means higher accuracy, up to 99.5% at Self Aware.[1]
A simulation costs energy, which depends on the model type. It also needs a catalyst item, which is usually a Prediction Matrix.[1]
Crafting Shaped
| Empty slot | any glass panes | Empty slot |
| any obsidian | ||
| any lapi gems | any lapi gems |
Recipe id: hostilenetworks:sim_chamber
Crafting Shaped
| any iron ingot | any glass panes | Empty slot |
| any glass panes | any glass panes | |
| Empty slot | any glass panes | any gold ingot |
Recipe id: hostilenetworks:prediction_matrix
The mod's screens describe two simulation modes. In training mode, simulations increase model data but produce no outputs. In inference mode, simulations produce output but the models gain no data.[4]
The generalized predictions form a chain: Generalized Overworld Prediction, Generalized Nether Prediction and Generalized Ender Prediction. There are crafting recipes for stepping up from Overworld to Nether and from Nether to Ender.[4][5]
Crafting Shaped
| Empty slot | Empty slot | |
| Empty slot | Empty slot |
Recipe id: hostilenetworks:living_matter/overworld_to_nether
Crafting Shaped
| Empty slot | Empty slot | |
| Empty slot | Empty slot |
Recipe id: hostilenetworks:living_matter/nether_to_ender
Loot Fabricator and Fabrication Directive
The Loot Fabricator uses power to turn predictions into real items. The available drops come from a pre-defined list, and you pick a target for each type of model. It remembers your choice for each type, so one fabricator can take several types of predictions.[1][4]
Crafting Shaped
| Empty slot | any netherite ingot | Empty slot |
| any diamond gem | any obsidian | any diamond gem |
| any gold ingot | any gold ingot |
Recipe id: hostilenetworks:loot_fabricator
A Fabrication Directive stores the selections from a Loot Fabricator. Right-click to copy, and Shift-Right-Click to apply. It can clone a fabricator's layout, which saves you setting up each machine by hand.[4]
Crafting Shaped
| Empty slot | any generalized prediction | Empty slot |
| Empty slot | Empty slot |
Recipe id: hostilenetworks:fab_directive
The Data Center
The Data Center is the multiblock endgame version of all this. It runs up to 25 Data Models at once, and each one produces one drop per cycle into a shared output buffer.[4]
Crafting Shaped
| any obsidian | ||
| any obsidian | any obsidian | any obsidian |
Recipe id: hostilenetworks:data_center
Insert Data Models into the 5×5 grid and matching items into the four input slots. Each active slot uses one input per cycle and produces the drop chosen by its Fabrication Directive.[4]
Building the shell
- Build a hollow 7×7×7 shell around the Data Center.[4]
- The floor must be obsidian. The walls and ceiling must be black stained glass.[4]
- The controller replaces one bottom-wall block.[4]
IO Ports and automation
A Data Center IO Port can replace any wall or ceiling block in the shell, and the controller detects it automatically. Right-click it to cycle through four modes: Energy, Models, Inputs and Outputs. Energy accepts power, while Models, Inputs and Outputs expose the relevant item slots. That is how you feed power and items in and pull drops out.[4]
The pack has changed the IO Port recipe from the mod's default, so use the recipe shown here.[5]
Crafting Shaped
| any iron ingot | any iron ingot | |
| any redstone storage block | ||
| any iron ingot | any iron ingot |
Recipe id: hostilenetworks:data_center_io_port
Unverified The exact energy cost of each simulation and fabrication, and any config options for them, are not shown by the sources here.
How do you progress through Hostile Neural Networks?
5 parts
The Hostile Neural Networks quest chapter walks you through one clear loop: make a Data Model, level it by killing mobs, run simulations, then turn the results into loot. Follow the quests in order and you'll hit every machine in the right sequence.[3]
Step 1: Learn and model mobs
Two quests open the chapter. "Learning Mobs" quest hands you the Deep Learner. Right click it to open a HUD and store Data Models. Each Learner holds 4 Data Models, and it collects data as long as it's anywhere in your inventory. Multiple Learners all count kills.[3]
Crafting Shaped
| any obsidian | any obsidian | |
| any glass panes | ||
| any obsidian | any redstone dust | any obsidian |
Recipe id: hostilenetworks:deep_learner
"Modeling Mobs" quest covers the Model Framework. Right click the mob you want to study with the Model Framework and it turns into that mob's Data Model. Put that model in a Deep Learner and kill that mob with the Learner in your inventory.[3]
Crafting Shaped
| any redstone dust | any redstone dust | |
| any gold ingot |
Recipe id: hostilenetworks:framework
Step 2: Level up your Data Model
"Leveling Up Data Models" quest explains that Data Models start at the first tier, Faulty. You need to get a model to at least Basic before you can run simulations on it. The Deep Learner HUD and the model's tooltip both show how many kills you still need.[3]
Step 3: Simulate
"Simulating Deaths" quest unlocks the Simulation Chamber, the main machine of the mod. It needs power, a Data Model, and at least one Prediction Matrix. The power cost depends on the model and shows in its tooltip.[3]
Crafting Shaped
| Empty slot | any glass panes | Empty slot |
| any obsidian | ||
| any lapi gems | any lapi gems |
Recipe id: hostilenetworks:sim_chamber
Crafting Shaped
| any iron ingot | any glass panes | Empty slot |
| any glass panes | any glass panes | |
| Empty slot | any glass panes | any gold ingot |
Recipe id: hostilenetworks:prediction_matrix
Simulations have two modes. Training makes the Data Model gain data, but produces no predictions and costs more Forge Energy. Inference puts its result onto the Matrix without giving the model any data, and it is faster. Every simulation produces some kind of Generalized Prediction, and the type depends on where the mob comes from. If the simulation succeeds, you also get a Mob Prediction. A higher-tier Data Model means a higher success rate.[3]
Step 4: Generalized Predictions and loot
Three side quests ask for the Generalized Predictions: a quest in "Hostile Neural Networks" for Generalized Overworld Prediction, a quest in "Hostile Neural Networks" for Generalized Nether Prediction, and a quest in "Hostile Neural Networks" for Generalized Ender Prediction. They are described as making various Overworld, Nether and End materials. The pack also has recipes that convert one tier of Generalized Prediction into the next.[3][5]
Crafting Shaped
| Empty slot | Empty slot | |
| Empty slot | Empty slot |
Recipe id: hostilenetworks:living_matter/overworld_to_nether
Crafting Shaped
| Empty slot | Empty slot | |
| Empty slot | Empty slot |
Recipe id: hostilenetworks:living_matter/nether_to_ender
"Creating Loot" quest brings in the Loot Fabricator. Give it power and a successful Mob Prediction, then choose which item you want from that mob's Data Model. It remembers your choice, so you can automate it. The quest's examples are Ender Dragons for Dragon Eggs, Zombies for Iron and Withers for Nether Stars.[3]
Crafting Shaped
| Empty slot | any netherite ingot | Empty slot |
| any diamond gem | any obsidian | any diamond gem |
| any gold ingot | any gold ingot |
Recipe id: hostilenetworks:loot_fabricator
After Creating Loot, the chapter opens a long run of quests that each ask you to obtain a Prediction. Treat them as a checklist of the mobs you can model.[3]
Endgame: the Data Center
"The Data Center" quest is the final step. The Data Center is a 7x7x7 multiblock that combines the Simulation Chamber and the Loot Fabricator. It handles 25 Self Aware Data Models at once, with the same effects as Inference simulations. Instead of Mob Predictions it produces loot directly, using the same modes as the Loot Fabricator.[3]
Crafting Shaped
| any obsidian | ||
| any obsidian | any obsidian | any obsidian |
Recipe id: hostilenetworks:data_center
For the build, the floor must be Obsidian and the walls and ceiling must be Black Stained Glass. The controller can replace one bottom-wall block, and Data Center IO Port blocks can replace the glass.[3]
The pack changed the IO Port recipe from the mod's own, so use the recipe shown here rather than one you remember from another pack.[5]
Crafting Shaped
| any iron ingot | any iron ingot | |
| any redstone storage block | ||
| any iron ingot | any iron ingot |
Recipe id: hostilenetworks:data_center_io_port
Unverified The sources name only Faulty (the starting tier), Basic (the minimum for simulations) and Self Aware (what the Data Center handles). The other tiers and the kill counts for each are not covered here, so check a Data Model's tooltip in-game.
Which All the Mods 10: To the Sky quests use Hostile Neural Networks?
2 parts
All the quests for this mod sit in one chapter, Hostile Neural Networks. The first two quests unlock after an intro quest called "Welcome to Hostile Neural Networks!", and the rest branch out from there: collect data, simulate, turn predictions into loot, then build the big multiblock.[3]
Quest order
- "Learning Mobs" quest asks for the Deep Learner. Right click it to open a HUD and store Data Models. Each Learner holds 4 Data Models and collects kill data as long as it is anywhere in your inventory.[3]
- "Modeling Mobs" quest asks for the Model Framework. Right click a mob with it to turn it into that mob's Data Model, then put the model in a Deep Learner and kill that mob. Like Learning Mobs, it unlocks straight after the intro quest.[3]
- "Simulating Deaths" quest asks for the Simulation Chamber and a Prediction Matrix. It unlocks after "Learning Mobs" quest and "Modeling Mobs" quest.[3]
- "Leveling Up Data Models" quest asks for a Data Model. It also unlocks after Modeling Mobs and Learning Mobs. New models start at the Faulty tier, and you need to reach at least Basic before you can run simulations on them.[3]
- "Creating Loot" quest asks for the Loot Fabricator. It unlocks after Simulating Deaths.[3]
- Three untitled quests also unlock after Simulating Deaths. They ask for the Generalized Overworld Prediction (a quest in "Hostile Neural Networks"), the Generalized Nether Prediction (a quest in "Hostile Neural Networks") and the Generalized Ender Prediction (a quest in "Hostile Neural Networks").[3]
- After Creating Loot comes a long row of untitled quests, each asking for a Prediction. One example is a quest in "Hostile Neural Networks".[3]
- "The Data Center" quest also unlocks after Creating Loot. It asks for the Data Center, a black glass block (several glass types are accepted) and the Data Center IO Port.[3]
What the key quests explain
The Simulation Chamber quest covers the two simulation modes. Training makes the Data Model gain data, but produces no predictions, and it costs more Forge Energy. Inference outputs the result onto the Matrix without the model gaining data, and it is faster. Every simulation gives some kind of Generalized Prediction, and a successful one also gives a Mob Prediction. Higher tier models succeed more often.[3]
The Loot Fabricator turns a successful Mob Prediction into an item you choose from that mob's Data Model. It remembers your choice, so you can automate it. The quest gives Ender Dragons for Dragon Eggs, Zombies for Iron and Withers for Nether Stars as examples.[3]
The Data Center quest describes a 7x7x7 multiblock that combines the Simulation Chamber and the Loot Fabricator. It handles 25 Self Aware Data Models at once with the same effects as Inference simulations, and it produces loot directly instead of Mob Predictions. The floor must be Obsidian, and the walls and ceiling must be Black Stained Glass. The controller can replace one bottom-wall block, and IO Ports can replace Black Stained Glass.[3]
What changed in Hostile Neural Networks between versions?
Hostile Neural Networks 6.5.1 is the version in All the Mods 10: To the Sky 2.0.4. The mod has been in the pack since 1.0 or earlier.[2]
Unverified The details of what changed between Hostile Neural Networks releases could not be confirmed, because the mod changelog contents were not available. No pack changelog entries for this mod were available either, so any changes the pack itself made to it across releases are unverified.
Sources
- Hostile Neural Networks project page (modrinth) · official mod data
- All the Mods 10: To the Sky 2.0.4 mod list · pack file
- All the Mods 10: To the Sky 2.0.4 quests: "Hostile Neural Networks" · pack file
- hostilenetworks tooltips and descriptions (English lang file) · official mod data
- All the Mods 10: To the Sky 2.0.4 recipe dump: recipes making Hostile Neural Networks items · pack recipe dump
Where this guide comes from
Written from what All the Mods 10: To the Sky itself ships: its files and recipes, the in-game guidebooks and the mods' own pages. Every statement links to the source it came from, and the numbers are checked against the pack before publishing.
Guide last verified Oct 16, 2026.
Guides for other versions: 2.0.1 to 2.0.2 · 1.6 to 2.0.0 · 1.1.1 to 1.2.1 · 1.0 to 1.0.2
Report wrong info
Version in each release
25 releases, first in 1.0
| All the Mods 10: To the Sky | Version | Jar | Where it came from |
|---|---|---|---|
| 2.1.0 Oct 15, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.7 Oct 5, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.6 Sep 21, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.5 Sep 21, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.4 Sep 11, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.3 Sep 9, 2026 | 6.5.1 | HostileNeuralNetworks-1.21.1-6.5.1.jar | From the published release |
| 2.0.2 Apr 1, 2026 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 2.0.1 Mar 27, 2026 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 2.0.0 Feb 24, 2026 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.9.1 Feb 5, 2026 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.9 Feb 4, 2026 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.8 Dec 20, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.7 Nov 28, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.6.1 Oct 23, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.6 Oct 17, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.5.1b Oct 2, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.5 Sep 26, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.4 Sep 11, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.3.1 Sep 3, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.2.1 Aug 28, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.2 Aug 28, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.1.2 Aug 22, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.1.1 Aug 21, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.0.2 Aug 15, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
| 1.0 Aug 14, 2025 | 6.3.0 | HostileNeuralNetworks-1.21.1-6.3.0.jar | From the published release |
"From pack files" means the list came from the pack's own files; "from the published release" means ForgeBase read the jars of the release itself.
Items and blocks
14 items from the 2.1.0 recipe dump.
All the Mods 10: To the Sky quests (45)
Quests from the pack's FTB Quests book that ask for this mod's items.
Hostile Neural Networks (8)
- Creating Loot
- Learning Mobs (The Learning HUD)
- Leveling Up Data Models
- Modeling Mobs ("Gathering Data")
- Simulating Deaths
- The Data Center (No Generative AI was used to make this quest)
- A quest in "Hostile Neural Networks"
- A quest in "Hostile Neural Networks"
Comments and tips
Loading comments...