RLHF
Reinforcement Learning from Human Feedback — a training technique where human evaluators rank model outputs to train a reward model, which then guides the LLM to produce preferred responses. The technique that made ChatGPT conversational.
Related Terms
Related Articles
More AI Terms
AI Agent
An autonomous AI system that can plan, use tools, and execute multi-step tasks to achieve goals. Goes beyond simple chat: agents can browse the web, write code, query databases, and interact with APIs. The frontier of applied AI.
AI Cost Optimization
Strategies for reducing the operational cost of AI systems: prompt caching, model selection (use cheaper models for simple tasks), batching, output length control, and caching frequent queries. Can reduce costs 60-90% without quality loss.
AI Safety
The field focused on ensuring AI systems behave as intended and don't cause harm. Encompasses alignment (AI goals match human goals), robustness (resistance to adversarial attacks), interpretability (understanding AI reasoning), and governance.
Claude
Anthropic's family of large language models designed for safety and helpfulness. Known for longer context windows (200K tokens), strong reasoning, and lower hallucination rates. Claude 4 is the latest generation. The main competitor to GPT-4.
Computer Vision
AI that enables machines to interpret visual information from images and video. Applications include object detection, image classification, facial recognition, OCR, and medical imaging. Models: YOLO, ResNet, Vision Transformers.
Context Window
The maximum amount of text an LLM can process in a single interaction, measured in tokens. Claude offers 200K tokens, GPT-4o offers 128K. Larger windows enable processing entire documents but increase cost and latency.