Building Blocks (15)
Category
Maturity
Architecture Pattern (1)
Causal Reasoning (1)
Epistemic Infrastructure (1)
Inference Optimization (2)
FlashAttention
UbiquitousThe attention mechanism in AI models — the part that lets them understand how different words in a s…
Inference Optimization
Speculative Decoding
ProductionLarge AI models generate text one word at a time, which is slow because each word requires running t…
Inference Optimization
Knowledge Strategy (1)
Memory Mechanism (1)
Modularity Pattern (1)
Reasoning Strategy (1)
Safety Mechanism (2)
ROME / MEMIT (Rank-One Model Editing / Mass-Editing Memory In a Transformer)
ExperimentalImagine you could reach into someone's memory and change one specific fact — say, change their memor…
Safety Mechanism
Sparse Autoencoders (SAEs) for Interpretability
ExperimentalInside an AI model, information is stored in a tangled, overlapping way — like many radio stations b…
Safety Mechanism
Training Method (2)
Knowledge Distillation
ProductionThe most powerful AI models are enormous and expensive to run. Knowledge distillation is like having…
Training Method
RLHF / DPO (Reinforcement Learning from Human Feedback / Direct Preference Optimization)
ProductionA freshly trained AI is like a brilliant but unsocialized savant — it has immense knowledge but no s…
Training Method
Verifiability Mechanism (2)
Proof-of-Reasoning
TheoreticalZKML can prove that an AI ran correctly (like proving a calculator worked). But Proof-of-Reasoning g…
Verifiability Mechanism
Zero-Knowledge Machine Learning (ZKML)
ExperimentalImagine you could prove you used a real calculator to get an answer, without showing anyone the calc…
Verifiability Mechanism