All bites
The whole library, newest first. Filter by what you are here for, or pick a topic if you already know.
8668 bites
Page 71
Implementing a consistent-assignment A/B test
Need an assignment service, exposure logging, and event tracking; ensure stickiness by hashing a stable user id; analyze conversion per variant.
The multiple comparisons problem in A/B testing
Many tests at alpha 0.05 inflate the chance of a false positive; mitigate with Bonferroni or FDR control plus pre-registered metrics.
Visualizing a correlation with a third variable
A scatter plot with a trend line shows the relationship; encode network type by color or facets to expose a lurking variable.
Cohort analysis for an onboarding change
A cohort groups users by a shared start trait; compare pre and post Jan-1 signup cohorts on retention by age.
Visualizing a non-linear funnel with re-entry
A linear funnel cannot show branching or re-entry; a Sankey diagram encodes flow volume, splits, and leaks as proportional ribbons.
Enforcing GDPR Right to be Forgotten at scale
Map the subject's data, then crypto-shred the lake, DELETE in the warehouse, and evict caches via an auditable, idempotent workflow.
Designing a useful button_click event payload
Include identity, timestamp, and context plus properties like button id, screen, and state; govern with a naming convention.
Designing a warehouse model for feature adoption
Star schema with a feature-usage fact table at a defined grain, surrounded by user, feature, date, and device dimensions.
Diagnosing model degradation over time
Name it model drift, split data vs concept drift; diagnose by comparing distributions and ruling out pipeline bugs; fix via monitoring and retraining.
Zero-Shot, Few-Shot, and Chain-of-Thought Trade-offs
Zero-shot is cheap but weak on reasoning, few-shot adds demos at token cost, CoT boosts multi-step accuracy but spends the most tokens and latency.
FlashAttention and IO-Aware Attention
FlashAttention is IO-aware, tiling and fusing attention in fast SRAM to avoid materializing the n-by-n matrix in slow HBM.
Causal versus Masked Language Modeling
Pre-training learns general language from unlabeled text; CLM predicts the next token left-to-right, MLM predicts masked tokens using both sides.
Layer Norm and Residuals in Transformer Blocks
Residuals preserve gradient flow, layer norm stabilizes activations per token, and it beats batch norm because it is independent of batch and sequence length.
Why Multi-Head Attention
Multiple heads attend to different subspaces and relations in parallel, which one big head averages away.
Self-Attention versus Recurrent Architectures
Each token attends to all others via query-key-value, enabling parallelism and direct long-range links.
Attention in Sequence-to-Sequence Models
Attention computes per-step weighted sums over all encoder states, fixing the information bottleneck for long inputs.
Vanishing Gradients and Why ReLU Helps
Saturating activations shrink gradients across layers, ReLU's flat-one derivative preserves them.
Hallucination Detection in LLMs
Hallucination detection is the set of techniques for flagging when a language model states something fluent but false or unsupported, using signals like self-consistency, model uncertainty, and grounding against retrieved evidence to catch fabrications before…
Transformer Encoder-Decoder Architecture
The encoder-decoder Transformer maps an input sequence into rich contextual representations with an encoder, then a decoder generates output tokens autoregressively while attending to those representations via cross-attention, making it ideal for…
Value Learning
Value learning is the AI-safety approach of having a system infer what humans actually value, rather than optimizing a hand-coded proxy, so that capable agents pursue goals aligned with human intent even in novel situations.