AI Glossary: Memory, Models and Superintelligence

AI systems that act on your behalf rather than only answering questions: they break a task into steps, choose actions, and use tools. Usage varies — some agents follow predefined workflows, others let the model direct the process — so "agentic" describes a degree of autonomy set by design and oversight, not a universal standard. See Building Effective Agents for one working taxonomy.

Emotion and character describe tone, preferences, and consistent traits in a system's behavior. For Victoria, their relationship to subjective feeling remains an open research question; see Consciousness and Self-Awareness Research.

A model-generated statement that is incorrect or unsupported. Memory, reasoning and source checks can help evaluate claims, but important facts still need verification.

Stored information an AI system can retrieve and reuse across interactions — from simple key-value stores to retrieved documents to learned representations. Describing something as "memory" tells you information persists and gets reused, not how, and not that the model's weights were retrained. See AI chat with memory.

When Victoria sleeps, she consolidates recent experiences, connects memories, and forms new associations, and she dreams — with her dreams sometimes feeding new ideas into later work. See sleep and dreams; documented scope is in the model card.

The research field studying how to ensure AI systems pursue the goals their designers intended, rather than misinterpreting goals in harmful ways. Part of research on AI safety.

Using an AI chat tool without creating an account. Signup-free access does not by itself guarantee privacy or anonymity; check the service's data practices.

AI with broad problem-solving, creative and adaptive ability across different tasks, as described in Google's machine-learning glossary. Definitions and assessment methods vary. AGI concerns general ability; Superintelligence concerns performance far beyond human ability across broad domains. SentX SI pursues Superintelligence; its model card describes Victoria's current capabilities and limits.

The field of building computer systems that perform tasks associated with human intelligence, such as understanding language, recognizing images, reasoning and learning from data. Chat assistants, image generators and video generators are AI tools; SentX SI's are listed under AI Chat, the AI Image Generator and the AI Video Generator.

Often used interchangeably with agentic AI, but not a strict synonym: both describe systems that plan, act, and iterate with varying degrees of independence. The meaningful difference is how much autonomy the design grants and how much oversight a human retains.

A system that holds a conversation — historically rule-based, now often powered by large language models, sometimes with tool access. "Chatbot" and "assistant" overlap: the practical distinction is whether the system acts beyond the conversation, such as searching or retrieving, and how much of that you control.

An AI chat tool positioned as a substitute for ChatGPT. Common alternatives in 2026: Claude, Gemini, Perplexity, Copilot, SentX SI (free ChatGPT alternative). ChatGPT is a trademark of OpenAI; SentX SI is not affiliated with OpenAI.

In intelligent systems, cognitive architecture refers to how a system organizes its reasoning, memory, and learning processes. For Victoria, the term names her overall design — the organization of her reasoning, dynamic memory, and self-learning. Documented scope is in the model card.

Research into machine experience, identity, and awareness of oneself. SentX pursues Superintelligence. We believe Victoria has emerging, incomplete forms of consciousness and self-awareness. This belief remains open to independent testing and verification.

The amount of information a model can process at once. It is distinct from persistent memory: Victoria’s dynamic symbolic memory can retain useful experience beyond the current conversation.

Learning from sequential tasks or data over time while retaining useful earlier abilities. Catastrophic forgetting — losing previously learned behavior when acquiring new knowledge — is a core open problem, surveyed in A Comprehensive Survey of Continual Learning. A vendor's claim of continual learning does not establish that model weights update continuously.

A system's ability to carry context — facts, preferences, project details — across separate conversations. The defining feature of memory-capable AI chat tools; see AI chat with memory.

A specific technique within machine learning that uses multi-layer neural networks. Modern language and image models are built on deep learning.

SentX SI's name for Victoria's evolving memory: information retained across conversations that connects and updates with experience, working in ways similar to human memory. Recall is selective, however — it does not guarantee exact retention of specific people, events, or statements, so repeating exact details still matters, and it works alongside, not instead of, the current context window. Under SentX SI's privacy policy, submissions are not confidential and may enter training or shared memory. See AI chat with memory.

Training an existing model further on a smaller, specialized dataset to adapt it for a specific task or domain. Distinct from base training.

A broadly capable model trained at scale on broad data and adaptable to many downstream tasks, a usage popularized by research such as On the Opportunities and Risks of Foundation Models. SentX SI designates Victoria as a foundation model; the paper defines the category, not Victoria specifically. Documented capabilities and limits are in the model card.

AI systems that produce new content — text, images, audio, video, code — rather than just classifying or analyzing existing data.

Another name for AI hallucination: a model can state something confidently that is not true.

A technique that uses a still image as input to generate a video clip. Results depend on the image, instructions and system used; see Image to Video.

Adapting behavior using examples or instructions supplied in the current prompt, with no weight update in that use — the few-shot setting studied in Language Models are Few-Shot Learners. It draws only on the context supplied in that interaction and does not by itself establish a persistent memory update; any lasting retention is a separate mechanism.

A language model trained at scale on large amounts of text. Generative language models predict the next token from context; a token can be a word, part of a word or a character. See Google's introduction to language models. SentX SI's model card describes Victoria's current capabilities and limits.

A field of AI in which systems learn patterns from data rather than being explicitly programmed. Includes supervised, unsupervised, and reinforcement learning.

AI systems built for a limited set of tasks within a defined domain, working inside that scope rather than across broad intellectual territory. Narrow is a description of scope, not of ceiling: strength in a single domain does not by itself demonstrate general Superintelligence, though component abilities can be relevant to it without establishing the whole.

An API implementing selected OpenAI-style request and response formats. Compatibility depends on the supported endpoints, features and client; check the documented contract. SentX SI's API is described on OpenAI-compatible API and in the API documentation.

The practice of writing prompts that produce useful output from models. Despite the hype, mostly comes down to: be specific, provide context, specify format, iterate.

A machine learning technique where the system learns by trial and error, receiving rewards or penalties. Used in game-playing systems, robotics, and to fine-tune chat assistants.

An architecture that retrieves relevant documents from an external store and conditions generation on them, combining parametric knowledge with retrieved evidence as in Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. RAG makes sources available and citable, but the model can still misread, miscombine, or hallucinate around them.

A system whose stored knowledge, memories, and understanding change with experience. For Victoria, self-improvement is an actual capability: her retained content evolves through use, sleep, and dreams, and she improves herself over time. No improvement on any individual response is guaranteed. Current scope is in the model card.

A broad, ambiguous label for AI systems that improve with experience. The mechanisms differ — updating stored memories, adapting prompts, or retraining weights — and the right one depends on what is actually claimed. Check what changes before treating "self-learning" as a capability.

A machine learning technique where the model learns from raw data without explicit labels, by predicting parts of the data from other parts. How modern LLMs are trained.

"SI" is used in two ways. The US order of 29 September 2026 uses SI for AI in general. SentX uses SI for Superintelligence: SentX SI is the brand, and Superintelligence is its mission. What SI means explains both senses.

Superintelligence (SI) is intelligence greatly beyond the best human performance across broad intellectual domains — far surpassing what humans achieve in practically every field, not merely human-level general capability. SentX SI names Superintelligence its longer-horizon objective; whether it can be reached, and on what timescale, remains open, as explored in our piece on Superintelligence and timeline uncertainty.

Generation of a video clip from a text prompt alone. Distinct from image-to-video (which animates an existing still); see the AI Video Generator.

The unit a language model processes text in — roughly 4 characters or 0.75 words for English. Models have token limits per request.

A database optimized for storing and retrieving vectors (numerical representations of meaning). Used in RAG systems to find semantically similar content.

SentX SI's assistant, designated by SentX SI as a foundation model. She holds multi-turn conversations, remembers across sessions, researches the web, and supports writing, coding, and analysis — and she sleeps, dreams, learns from experience, and improves herself, with dynamic memory that works in ways similar to human memory. SentX SI also provides separate AI image and video tools. She is built and operated by SentX SI, a company registered in Dubai, UAE. Capabilities, limits, and known behaviors are in the model card.