Seattle Skeptics on AI

Tamara Weed, Apr, 5 2026

Explore the hidden privacy and security risks of distilled LLMs. Learn why model compression doesn't stop PII leaks and how to use Intel TDX to secure your AI deployment.

Tamara Weed, Apr, 4 2026

Learn how Federated Learning enables training Large Language Models (LLMs) across decentralized data sources to ensure privacy and bypass data centralization.

Tamara Weed, Apr, 1 2026

Exploring emergent capabilities in Generative AI: definition, examples like chain-of-thought, the 'mirage' debate, and safety implications for 2026.

Tamara Weed, Mar, 31 2026

Explore how layer dropping and early exit techniques accelerate Large Language Model inference, reducing latency and costs without sacrificing accuracy.

Tamara Weed, Mar, 30 2026

Explore the distinct roles of API Gateways and Service Meshes in modern microservices architecture, including performance comparisons and implementation strategies for 2026.

Tamara Weed, Mar, 29 2026

Explore why AI hallucinations happen and learn practical strategies like RAG and RLHF to reduce factual errors in generative systems.

Tamara Weed, Mar, 28 2026

Traditional metrics like BLEU fail to capture LLM meaning. Learn why semantic metrics like BERTScore and LLM-as-a-Judge provide accurate quality assessment for modern AI deployments.

Tamara Weed, Mar, 27 2026

Discover how vibe coding transforms global team productivity by turning natural language into executable code. Learn about real-world use cases, velocity gains, and infrastructure needs.

Tamara Weed, Mar, 26 2026

Learn how positional encoding solves the word order problem in Transformers. We explore absolute, relative, and rotary methods, recent research findings, and future trends.

Tamara Weed, Mar, 25 2026

Explore how Large Language Models transform enterprise knowledge management by turning static documents into dynamic Q&A systems. Learn about RAG architecture, security challenges, and implementation costs.

Tamara Weed, Mar, 23 2026

Learn how memory planning techniques like CAMELoT and Dynamic Memory Sparsification reduce OOM errors in LLM inference by 40-60% without sacrificing accuracy - and why quantization alone isn't enough for long-context tasks.

Tamara Weed, Mar, 23 2026

Memory planning techniques like CAMELoT and Dynamic Memory Sparsification let LLMs handle long contexts without OOM crashes-cutting memory use by 50% while improving accuracy. No more brute-force GPU scaling needed.