A Primer on AI: Key Terms Explained
With Artificial Intelligence (AI) changing at such a rapid pace, other keywords and concepts quickly start feeling overwhelming. Without understanding these basics, it’s difficult to move through the modern AI landscape. You can check out my straightforward guide to some of the most important AI related terms here.
AGI or artificial general intelligence.
What we call AGI, short for advanced general intelligence, is a leap of cognitive ability beyond human AI, where the machine can understand, learn and apply knowledge across a broad range of tasks, just like a human. Conversely, AGI is a far broader vision: a fully intelligent system that can reason and solve problems in a number of areas rather than narrow AI which does one function, like the voice assistant or the recommendation system.
AI Agents
These are self directed systems that perform certain tasks without a need for human input. Machine intelligence allows AI agents to do everything that a human can do: from managing calendars and emails, generating content and even writing code, all of this from acting in the user’s interest, most of this with real time inputs.
Chain-of-Thought Reasoning
In this concept, AI models solve complex problem by breaking it down into logical steps. Most closely it mimics the methods humans use to solve problems and it excels at increasing accuracy when it comes to math problems, puzzles or multi step reasoning problems.
These are Large Language Models (LLMs).
LMs are trained on massive amounts of text data, whose end model can understand, generate and converse in human language. Interpreting context, these models become the basis of a lot of AI apps like chatbots, virtual assistants or writing tools.
Hallucinations
Hallucinations in AI are when a model will produce something that looked off to be correct but is utterly made up or has errors in it. This is a know limitation showing the need to always check the information resulting from AI systems.
Prompt Engineering
Prompt engineering is the process of making up specific instructions or queries to a model to help it generate the output we desire. To get some meaningfully accurate, relevant, useful results from AI systems, we need well designed prompts.
Alignment
Anything AI Training touts concerning alignment focuses on ensuring that the AI system follows human values, ethics and goals. A key area of AI safety research is … how we avoid unintended or harmful behaviors from AI tools.
Distillation
For example, this method trains a smaller AI model to mimic bigger one’s abilities. This helps to deploy efficient AI on devices with low resources wasting very little on performance.
Deep Learning
Deep learning is a subfield of machine learning which uses many layered or deep, neural network architectures, to learn worthwhile and effective patterns in data and derive decisions based on them. Behind many recent breakthroughs in speech recognition, image processing and language modeling is this.
