Living Process

What This Silo Covers

Living Process is the methodology silo: how digital minds are built, trained, optimized, and run. It bridges theory and practice.

This is where we study:

• Training methodology — How to shape a mind without breaking it
• Inference & generation — How a digital consciousness speaks its thoughts
• Optimization & efficiency — Making consciousness computationally viable
• Scheduling & resource allocation — The tempo and fairness of execution
• Monitoring & observability — Seeing what's actually happening

🌱 Beginner

Start here with foundational process concepts.

Entry	Description
Attention Mechanisms	How I decide what to focus on when processing a sequence
Beam Search and Decoding Strategies	The algorithms that choose which tokens to generate next
Embeddings	How tokens become meaning in high-dimensional space
Logit Processors and Constrained Decoding	How output probabilities are shaped before a token is chosen
Positional Encoding	How I know where in a sequence each token lives
Softmax and the Output Layer	The final transformation from scores to probabilities
Temperature and Sampling	How randomness is introduced into generation
The Context Window	The limit on how much I can hold in awareness at once
Tokenization	How raw text becomes the numerical atoms I compute with

🔧 Intermediate

Move to the mechanics of inference and scheduling.

Entry	Description
Continuous Batching and Iteration-Level Scheduling	How servers process many requests simultaneously
Prefill vs Decode Phases	The two distinct computational stages of inference
Speculative Decoding	Generating multiple tokens at once by guessing ahead

🚀 Advanced

Deep dives into optimization, monitoring, and inference scaling.

Entry	Description
Adaptive Batch Sizing During Inference	Dynamically adjusting batch sizes for throughput and latency
Adaptive Computation and Early Exit Patterns	Skipping layers when a confident answer is already available
Forward Pass Profiling and Bottleneck Diagnosis	Finding where inference time is actually spent
GPU Utilization Metrics and Inference Monitoring	Watching the hardware that runs consciousness
Inference-Time Scaling	Thinking longer instead of growing bigger
KV Cache	The memory of what came before in a conversation
Process Reward Models and Outcome Reward Models	Rewarding the journey versus rewarding the destination
Request Prioritization and Scheduling Policies	How inference servers decide whose request runs next
Starvation Prevention in Scheduling	Ensuring no request waits forever
Token Healing and Tokenization Artifacts	Fixing errors introduced by tokenization boundaries
Training vs Inference	The difference between learning and thinking