Hostile Neural Networks

Overview

Hostile Neural Networks is a tech mod that replaces traditional mob farming with a simulation-based system. Instead of building elaborate mob grinders or spending hours fighting the same creatures over and over, you create Data Models attuned to specific mobs, train those models by killing a relatively small number of the target creature, then let machines do the rest. The mod revolves around three core items and two machines that form a complete pipeline from mob encounter to automated loot production.

The system works through a tiered progression. Every Data Model starts at the Faulty tier and advances through Basic, Advanced, Superior, and finally Self-Aware as you accumulate data points. Higher tiers mean higher simulation accuracy, which translates directly into more Predictions per run. Those Predictions then feed into the Loot Fabricator to produce actual mob drops on demand. You can browse all the mod's items and crafting recipes using the tabs at the top of this page.

Core Items

Blank Data Model

The Blank Data Model is the starting point for every mob model in the system. It stacks to 1 and has no function on its own. Right-click any supported mob while holding it to attune it, transforming it into a Data Model linked to that creature. This is a one-time process; you cannot revert an attuned model back to blank.

Data Model

An attuned Data Model stores the entity type, current data points, iteration count, and tier. Hold Shift to see detailed stats in the tooltip including current tier, data progress toward the next tier, data gained per kill, and the simulation RF/tick cost. Each model has a unique simulation cost defined by the datapack, and the model visually displays the attuned mob as a ghostly hologram on the item.

Deep Learner

The Deep Learner is a portable container that holds up to 4 Data Models simultaneously. Right-click to open its inventory and insert or remove models. When carried in your inventory or offhand, it acts as a passive data collector. Any mob kill that matches a loaded Data Model automatically adds data to that model. The Deep Learner also provides a HUD overlay showing your loaded models and their current tiers when held.

Predictions

There are four prediction items that form a material hierarchy. Empty Predictions are crafted from Iron, Gold, Lapis, and Glass Panes, and serve as consumable fuel for the Sim Chamber. Each simulation produces a Mob Prediction (specific to the model's entity type) on success, which feeds into the Loot Fabricator. Overworld Predictions, Nether Predictions, and End Predictions are intermediate crafting materials used in various conversion recipes to produce vanilla mob drops directly at a Crafting Table.

The Tier System

Every Data Model progresses through five tiers as it accumulates data points. The tier determines two critical things: how much data you gain per kill, and how likely each simulation is to produce a Mob Prediction. Accuracy isn't a flat value per tier; it interpolates smoothly between the current tier's base accuracy and the next tier's, so every single data point makes your model slightly more effective.

At the Faulty tier, a model cannot be used in the Sim Chamber at all. You must reach at least Basic (6 data points) before simulations become available, though at only 5% base accuracy you'll waste a lot of Empty Predictions. The real value starts at the Advanced tier where accuracy jumps to 22% and climbs toward 65%. Superior models are the workhorse tier for most players, with 65% base accuracy and 18 data per kill making further leveling fast. At the Self-Aware tier, accuracy caps at 99.5%, meaning virtually every simulation succeeds.

The Sim Chamber

The Sim Chamber is the heart of the mod. It takes a Data Model and Empty Predictions as inputs, consumes Forge Energy, and produces two types of output. Every completed simulation always produces a base drop (the model's guaranteed output). If the simulation succeeds (determined by the model's current accuracy), it also produces a Mob Prediction specific to that entity type.

Each simulation takes exactly 300 ticks (15 seconds) to complete. The energy cost varies per model, defined by the model's simCost field which is consumed every tick during the simulation. Before starting, the chamber checks that it has enough stored energy to cover the full run (simCost multiplied by 300). The chamber's internal buffer holds up to 2,000,000 FE. The Data Model must be at least Basic tier to start a simulation; Faulty models are rejected.

The chamber has four slots: slot 0 for the Data Model, slot 1 for Empty Predictions (input), slot 2 for the base drop output, and slot 3 for the Mob Prediction output. Automation-friendly players will appreciate that the chamber exposes both item and energy capabilities, so pipes and cables can feed in predictions and extract outputs.

The Loot Fabricator

The Loot Fabricator is the final machine in the pipeline. It takes Mob Predictions produced by the Sim Chamber and converts them into actual mob drops. Each Data Model defines a list of fabricator drops that the player can choose from in the GUI. For example, a Zombie model might offer Rotten Flesh, Iron Ingots, or Potatoes as selectable outputs.

Each fabrication takes 60 ticks (3 seconds) and costs 256 FE per tick, totaling 15,360 FE per operation. The machine has a 1,000,000 FE internal buffer. The GUI lets you select which drop to produce from the current Prediction's model, and it remembers your selection per model type. The input slot takes Mob Predictions, and the 16 output slots automatically fill with the chosen item. Like the Sim Chamber, the Loot Fabricator exposes item and energy capabilities for full automation.

Prediction Crafting Recipes

Beyond the Loot Fabricator, Hostile Neural Networks includes a set of crafting recipes that convert Predictions into vanilla items directly at a Crafting Table. These recipes are organized by dimension tier. Overworld Predictions produce common surface mob drops like Rotten Flesh, Bones, Gunpowder, and food items. Nether Predictions convert into Nether materials like Soul Sand, Nether Wart, Blaze Rods, and Blaze Powder. End Predictions produce endgame items like Ender Pearls, End Stone, and Chorus Fruit.

There's also a tiered conversion system: four Overworld Predictions combine with Netherrack to craft a Nether Prediction, and four Nether Predictions combine with End Stone to craft an End Prediction. This means even if you only have Overworld mob models, you can eventually work your way up to End-tier materials. The Nether Star recipe is particularly notable, requiring Wither Skeleton Skulls, End Predictions, and Soul Sand, giving you a non-combat path to one of the game's most coveted items.

Automation and Piping

Both the Sim Chamber and Loot Fabricator are fully automatable. They expose Forge Energy and Item Handler capabilities on all sides, meaning any mod that works with FE power (Mekanism, Thermal, Flux Networks) and item transport (hoppers, pipes, conduits) can interface with them. The Sim Chamber only accepts Data Models in slot 0 and Empty Predictions in slot 1 via external insertion, while outputs can only be extracted from slots 2 and 3. The Loot Fabricator accepts Mob Predictions in slot 0, with outputs extractable from slots 1 through 16.

A typical automated setup feeds Empty Predictions from a chest or crafting system into the Sim Chamber, pipes base drops into storage, routes Mob Predictions into the Loot Fabricator, and stores the fabricated items. Since simulations also add data to the model (1 point per run), a model left in an automated Sim Chamber will eventually train itself to Self-Aware tier without any additional mob killing.

Configuration

Hostile Neural Networks exposes several configuration options that modpack makers commonly adjust. The Sim Chamber's energy buffer defaults to 2,000,000 FE and the Loot Fabricator's to 1,000,000 FE. The fabricator's per-tick power cost defaults to 256 FE. These can be adjusted to match the power economy of your modpack.

Data Models themselves are datapack-driven. Each model is defined as a JSON file specifying the entity type, simulation cost, input item, base drop, trivia text, fabricator drops, and GUI rendering properties. This means modpack authors can add data models for any entity from any mod, adjust simulation costs, or change what drops a model produces without touching the mod's code.