Artificial Intelligence (AI)
Systems that perform tasks requiring human-like perception, reasoning, or language.
Tech Insights
Top 50 AI & silicon buzzwords
Essential notes on AGI, ASI, NVIDIA, Cerebras, Intel, Etched, Apple MLX, deep learning, and the vocabulary shaping enterprise AI strategy.
Filter
AGI
Artificial General Intelligence—models with broad, human-level capability across domains.
ASI
Artificial Superintelligence—hypothetical AI surpassing human performance in virtually all areas.
Deep Learning
Neural networks with many layers that learn hierarchical representations from data.
Machine Learning (ML)
Algorithms that improve through experience without explicit rule programming.
Large Language Model (LLM)
Transformer-scale models trained on vast text for generation and understanding.
Transformer
Attention-based architecture powering modern LLMs and multimodal systems.
Neural Network
Interconnected layers of nodes that approximate complex functions via training.
RAG
Retrieval-Augmented Generation—grounding LLM answers in external knowledge bases.
RLHF
Reinforcement Learning from Human Feedback—aligning models with human preferences.
Fine-Tuning
Adapting a pre-trained model to a specific task or domain with additional data.
Inference
Running a trained model to produce predictions or generated content.
Training
Optimizing model weights on datasets—compute-intensive pre-deployment phase.
Multimodal AI
Models processing text, images, audio, and video in unified architectures.
Embeddings
Dense vector representations capturing semantic meaning for search and clustering.
NVIDIA
Dominant GPU supplier for AI—CUDA ecosystem, H100, and Blackwell data-center chips.
CUDA
NVIDIA's parallel computing platform—the de facto standard for GPU ML workloads.
H100
NVIDIA Hopper GPU—workhorse for large-scale LLM training and inference clusters.
Blackwell
NVIDIA's next-gen architecture targeting trillion-parameter-class AI infrastructure.
Cerebras
Wafer-scale AI chips (WSE)—dinner-plate-sized processors for massive models.
WSE
Cerebras Wafer Scale Engine—single wafer die with millions of cores for AI.
Intel
CPU and accelerator vendor—Gaudi AI chips and data-center Xeon platforms.
Intel Gaudi
Intel's AI accelerator line designed for training and inference efficiency.
Etched
AI chip startup building transformer-specific ASICs (Sohu) for inference at scale.
ASIC
Application-Specific Integrated Circuit—custom silicon optimized for one workload.
GPU
Graphics Processing Unit—parallel processors essential for deep learning throughput.
TPU
Google's Tensor Processing Unit—ASIC family built for TensorFlow and JAX workloads.
Apple MLX
Apple's array framework for efficient ML on Apple Silicon—NumPy-like API on M-series chips.
PyTorch
Meta's dynamic deep-learning framework widely used in research and production.
TensorFlow
Google's end-to-end ML platform for training, deployment, and edge inference.
JAX
Google's composable transformations for high-performance numerical computing and ML.
ONNX
Open Neural Network Exchange—interoperable format for moving models across runtimes.
vLLM
High-throughput LLM inference engine with PagedAttention for serving at scale.
LangChain
Orchestration framework for chaining LLMs, tools, and retrieval pipelines.
Hugging Face
Hub and libraries for open models, datasets, and transformers ecosystem.
OpenAI
Frontier lab behind GPT models—APIs, ChatGPT, and enterprise AI partnerships.
Anthropic
AI safety–focused lab—Claude models and constitutional AI research.
ChatGPT
OpenAI's conversational product—GPT-4 class models via chat interface.
Gemini
Google DeepMind multimodal LLM family—text, code, image, and video.
Qwen
Alibaba's open-weight LLM series—strong multilingual and coding performance.
DeepSeek
Chinese AI lab—efficient MoE models and competitive reasoning benchmarks.
Foundation Model
Large pre-trained model adaptable via fine-tuning to many downstream tasks.
Prompt Engineering
Crafting inputs to steer LLM behavior, quality, and reliability.
AI Safety
Research and practices to reduce harm, misuse, and misalignment in AI systems.
Alignment
Ensuring AI systems pursue intended goals and human values.
Hallucination
When models generate plausible but false or unsupported information.
Token
Subword unit of text—LLMs process and bill usage by token count.
Context Window
Maximum tokens a model can consider in a single request or conversation.
Quantization
Reducing model precision (e.g. INT8/FP4) to cut memory and speed inference.
Mixture of Experts (MoE)
Sparse architectures activating subsets of parameters per input for scale.
Synthetic Data
AI-generated training data to augment or replace scarce real-world datasets.
Edge AI
Running models on devices (phones, IoT) rather than centralized cloud only.
MLOps
Operational practices for deploying, monitoring, and versioning ML in production.
AI Act
EU regulatory framework governing high-risk and general-purpose AI systems.