A Quick Guide to AI Essentials
AI is infiltrating every aspect of modern life, from music and media to business, productivity, and even dating. Keeping up can be tough — so here’s your guide to the latest developments, essential terms, and top AI companies to know to stay informed in this rapidly evolving field.
What is AI?
Artificial intelligence, also known as machine learning, is a type of software system rooted in neural networks. While the technique behind AI was pioneered decades ago, recent advancements in computing power have spurred its growth. AI now powers voice and image recognition, generates synthetic images and speech, and is evolving to handle tasks like web browsing, ticket booking, and even recipe tweaks.
For those concerned about a Matrix-style machine uprising — don’t worry; we’ll address that later! Our AI guide includes three main parts, which we’ll update regularly, and can be read in any order:
- Key concepts in AI, from foundational to cutting-edge.
- An overview of major players in the AI industry and why they matter.
- A curated list of recent headlines and developments in AI.
By the end, you’ll be as up-to-date as possible in the world of AI. This guide will be expanded over time as AI continues to advance. Though core AI concepts date back more than 50 years, only recently have many of them gained familiarity. So if AI feels overwhelming, you’re not alone.
One clarification: while we call it “artificial intelligence,” that term can be misleading. AI systems are closer to advanced calculators than true brains, offering flexible outputs based on input. Think of it as “artificial coconut” — it’s imitation intelligence.
With that, here are essential AI terms:
Neural Network
The brain contains interconnected neurons forming complex networks that store information and perform tasks. Software-based neural networks have been attempted since the 1960s, but only recently, with the rise of GPUs, have they thrived. Neural networks can quickly process inputs through a complex statistical web, generating outputs. This system is called a model.
Model
A model is a collection of code that accepts inputs and produces outputs, similar to a statistical or simulation model. Models vary in size, storage, and power requirements, depending on how they’re trained.
Training
To create an AI model, the neural network is exposed to data in a dataset or corpus, forming a statistical representation of that data. Training is highly resource-intensive, taking weeks on high-powered computers, depending on the dataset’s size. However, once trained, the model can be used efficiently, a process known as inference.
Inference
When an AI model operates, this is called inference — concluding by reasoning through available evidence. For example, if asked, “Complete the sequence: red, orange, yellow…”, the AI finds this matches a familiar pattern and continues it.
Generative AI
Generative AI refers to AI that produces new outputs, like images or text. It’s widely popular but keep in mind — AI output isn’t necessarily factual, even when it’s convincing.
Large Language Model (LLM)
Trained on vast text datasets, LLMs are powerful AI models that communicate in natural language. ChatGPT, Claude, and LLaMa are examples of LLMs. While they’re impressive, they are essentially pattern recognition systems and may “hallucinate” responses.
Foundation Model
Foundation models are extensively trained and computationally intensive, but can be “trimmed” to fit smaller systems through parameter reduction.
Fine-tuning
Generalist models like GPT-4 can be fine-tuned for specific tasks by further training on specialized datasets, allowing for better domain-specific responses without losing general knowledge.
Reinforcement Learning from Human Feedback (RLHF)
RLHF is a type of fine-tuning where human feedback is used to improve an LLM’s communication skills.
Diffusion
This is a popular method for image generation, used by Stable Diffusion and others. It involves AI models learning to reverse digital noise to create images.
Hallucination
In AI, “hallucination” refers to the generation of incorrect or invented information, which can be problematic in factual contexts.
AGI or Strong AI
Artificial General Intelligence (AGI) refers to AI that not only performs tasks but learns and improves independently. While this concept is still far from reality, it continues to drive discussions about AI’s potential risks and benefits.
Top Players in AI:
1) OpenAI: Known for ChatGPT, OpenAI started as an open research organization and has since become a for-profit entity led by Sam Altman.
2) Microsoft: Early OpenAI investor, Microsoft powers Bing with OpenAI’s technology and conducts AI research across other applications.
3) Google: Known for pioneering transformers, Google is catching up in LLMs and AI-driven search improvements.
4) Anthropic: Founded by former OpenAI researchers, Anthropic aims to develop safe, ethical AI with its Claude models.
5) Stability: Known for open-source generative AI models, Stability’s approach allows anyone with hardware to run AI, stirring debates about ethical use.