AI Ecosystem
A brief list of more essential parts of the AI ecosystem you should get familiar with if you want to understand and even use AI properly.
Understanding what artificial intelligence (AI) is, or isn't, is a critical step to navigating the modern landscape of almost industry.
Don't be discouraged if you know little or if it's hard to understand at first. Remember Benjamin Franklin's own words, "The doorstep to the temple of wisdom is knowledge of our own ignorance."
AI List Work in Process (WIP)
- AI Agent: Software entity that uses an LLM as its core engine to autonomously perceive its environment, reason, plan, and execute multi-step actions to achieve goals. (Wikipedia)
- Anthropic: An AI safety and research corporation founded by former OpenAI executives, known for the Claude model family. (Wikipedia)
- Artificial General Intelligence (AGI): The theoretical milestone where an autonomous AI system possesses human-level or superior cognitive capabilities across a broad range of disciplines. (Wikipedia)
- Context Window: The maximum volume of text tokens an LLM can read, process, and retain in its memory during a single prompt and generation cycle. (Wikipedia)
- DeepSeek: An AI research company specializing in highly compute-efficient open-weight architectures, low-cost training methodologies, and advanced reinforcement learning frameworks. (Wikipedia)
- Fine-Tuning: The process of taking a pre-trained base model and training it further on a targeted dataset to adapt it for specific tasks or domains. (Wikipedia)
- GGUF: Binary file format optimized for rapid loading and execution of quantized models on CPU/GPU hardware. (Wikipedia)
- Hugging Face: Central registry and open-source platform for hosting, evaluating, and collaborating on ML models and datasets. (Wikipedia | Website)
- Intelligent Agent: The foundational computer science concept of an autonomous entity that directs its activity toward achieving goals based on environmental inputs. (Wikipedia)
- Large Language Model (LLM): Deep neural network architecture trained to predict subsequent text sequences. (Wikipedia)
- Llama (Meta): Meta's flagship family of high-performance, open-weights LLMs that catalyzed the open-source AI development community. (Wikipedia)
- llama.cpp: Lightweight C/C++ inference engine for executing open-weight models locally with minimal resource overhead. (GitHub)
- LoRA (Low-Rank Adaptation): An efficient fine-tuning technique that freezes base model weights and injects tiny, trainable rank-decomposition matrices to minimize computing overhead. (Wikipedia)
- Mistral AI: A French AI company recognized for pioneering highly efficient open-weight models, particularly Mixture of Experts architectures. (Wikipedia)
- Mixture of Experts (MoE): Architecture routing tokens to specialized neural sub-networks, scaling capacity without proportional compute overhead. (Wikipedia | HF Blog)
- OpenAI: The prominent AI research and deployment company behind the GPT series, ChatGPT, and the broader commercialization of generative AI. (Wikipedia)
- Quantization: Compression technique that reduces the numerical precision of model weights (e.g., 16-bit to 4-bit) to drastically decrease memory footprints for local deployment. (Wikipedia)
- Retrieval-Augmented Generation (RAG): Framework that optimizes LLM outputs by querying external, authoritative knowledge bases before generating a response. (Wikipedia)
- Tokens (Input & Output): Fundamental text units processed by an LLM. Input acts as the prompt; output is the generated prediction. (Wikipedia)
- Training Data (Common Crawl): Massive, raw web-scraped text corpora used for foundational pre-training. (Wikipedia)
- Transformer: The foundational deep learning architecture utilizing self-attention mechanisms that powers virtually all modern LLMs. (Wikipedia)
- Vector Database: Specialized database designed to store, index, and rapidly query high-dimensional vector embeddings, a critical component of RAG pipelines. (Wikipedia)
- vLLM: High-throughput serving engine utilizing PagedAttention for memory-efficient KV cache management. (Documentation)
