The practice of ensuring AI systems behave according to human intentions and values—being helpful, harmless, and honest. Alignment encompasses training-time techniques (RLHF, Constitutional AI, DPO), inference-time guardrails, and evaluation through red teaming. As models become more capable, alignment becomes critical to prevent harmful content generation or manipulation by bad actors.
Empieza por la explicación más corta y útil antes de profundizar.
The practice of ensuring AI systems behave according to human intentions and values—being helpful, harmless, and honest. Alignment encompasses training-time techniques (RLHF, Constitutional AI, DPO), inference-time guardrails, and evaluation through red teaming. As models become more capable, alignment becomes critical to prevent harmful content generation or manipulation by bad actors.
Usa primero la analogía corta para razonar mejor sobre el término cuando aparezca en código, docs o prompts.
Piensa en esto como una pieza de la pila de contexto o inferencia usada en productos con agentes o LLMs.
Ubica el término dentro de la capa de Solana en la que vive para razonar mejor sobre él.
LLMs, RAG, embeddings, inferencia y primitivas orientadas a agentes.
Convierte el término de vocabulario en algo operacional para producto e ingeniería.
Este término desbloquea conceptos adyacentes rápido, así que funciona mejor cuando lo tratas como un punto de conexión y no como una definición aislada.
Usa este bloque compacto cuando quieras dar contexto sólido a un agente o asistente sin volcar toda la página.
AI Alignment (ai-alignment) Categoría: IA / ML Definición: The practice of ensuring AI systems behave according to human intentions and values—being helpful, harmless, and honest. Alignment encompasses training-time techniques (RLHF, Constitutional AI, DPO), inference-time guardrails, and evaluation through red teaming. As models become more capable, alignment becomes critical to prevent harmful content generation or manipulation by bad actors. Aliases: AI Safety Relacionados: RLHF (Reinforcement Learning from Human Feedback), Constitutional AI, DPO (Direct Preference Optimization)
Usa contexto del glosario, relaciones entre términos, modelos mentales y builder paths para recibir respuestas estructuradas en vez de output genérico.
Opcional: pega código Anchor, Solana o Rust para que el Copilot mapee primitivas de vuelta al glosario.
El Copilot responderá usando el término actual, conceptos relacionados, modelos mentales y el grafo alrededor del glosario.
Estas ramas muestran qué conceptos toca este término directamente y qué existe una capa más allá de ellos.
A training technique that aligns LLM outputs with human preferences. Process: (1) train a reward model from human comparisons of outputs, (2) use reinforcement learning (PPO) to optimize the LLM against the reward model. RLHF makes models more helpful, harmless, and honest. Used by Claude, ChatGPT, and other assistants. Alternatives include DPO (Direct Preference Optimization) and Constitutional AI.
An alignment technique developed by Anthropic where an AI model is guided by a 'constitution'—a set of explicit principles defining allowed and disallowed behavior—rather than relying solely on human feedback. The model critiques and revises its own outputs against these principles. Constitutional Classifiers extend this by training input/output classifiers that detect policy violations at low compute cost.
A simplified alternative to RLHF that aligns LLM outputs with human preferences without training a separate reward model or using reinforcement learning. DPO directly optimizes a policy using pairs of preferred and dispreferred outputs, making it computationally cheaper and more stable than RLHF's multi-stage pipeline. Widely adopted in 2024-2025 for fine-tuning open-source models.
Estos son los siguientes conceptos que vale la pena abrir si quieres que este término tenga más sentido dentro de un workflow real de Solana.
Las entradas del glosario se vuelven útiles cuando están conectadas. Estos enlaces son el camino más corto hacia ideas adyacentes.
A training technique that aligns LLM outputs with human preferences. Process: (1) train a reward model from human comparisons of outputs, (2) use reinforcement learning (PPO) to optimize the LLM against the reward model. RLHF makes models more helpful, harmless, and honest. Used by Claude, ChatGPT, and other assistants. Alternatives include DPO (Direct Preference Optimization) and Constitutional AI.
An alignment technique developed by Anthropic where an AI model is guided by a 'constitution'—a set of explicit principles defining allowed and disallowed behavior—rather than relying solely on human feedback. The model critiques and revises its own outputs against these principles. Constitutional Classifiers extend this by training input/output classifiers that detect policy violations at low compute cost.
A simplified alternative to RLHF that aligns LLM outputs with human preferences without training a separate reward model or using reinforcement learning. DPO directly optimizes a policy using pairs of preferred and dispreferred outputs, making it computationally cheaper and more stable than RLHF's multi-stage pipeline. Widely adopted in 2024-2025 for fine-tuning open-source models.
Estas entradas viven junto al término actual y ayudan a que la página se sienta parte de un grafo de conocimiento más amplio en lugar de un callejón sin salida.
A neural network trained on vast text corpora to understand and generate human language. LLMs (GPT-4, Claude, Llama, Gemini) use transformer architectures with billions of parameters. They power chatbots, code generation, summarization, and reasoning tasks. In blockchain development, LLMs assist with smart contract writing, audit review, documentation, and code explanation.
The neural network architecture underlying modern LLMs, introduced in 'Attention Is All You Need' (2017). Transformers use self-attention mechanisms to process input sequences in parallel (unlike recurrent networks). Key components: multi-head attention, positional encoding, feedforward layers, and layer normalization. Variants include encoder-only (BERT), decoder-only (GPT), and encoder-decoder (T5).
A neural network component that allows models to weigh the relevance of different parts of the input when producing output. Self-attention computes query-key-value dot products across all positions, enabling each token to 'attend' to every other token. Multi-head attention runs multiple attention functions in parallel. Attention is O(n²) in sequence length, driving context window research.
A large AI model trained on broad data that can be adapted for many downstream tasks. Foundation models (GPT-4, Claude, Llama 3, Gemini) are pre-trained on internet-scale text/code and can be fine-tuned, prompted, or used via APIs for specific applications. The term emphasizes that one base model serves as the foundation for diverse use cases rather than training task-specific models.