[AI Terminology Illustrated Series] #2 Five AI Terms to Get You Ahead | AI x Information Edition
This time, we are covering AI x Information.
Basic AI terms were summarized in #1. This time, I will narrow it down to five terms that you often hear but might not fully understand, to take you a step further.
You won't have trouble if you don't know them, but knowing them will significantly increase your resolution when reading AI news.
We have completely revised the design this time🦉
1. RAG (Retrieval-Augmented Generation)
Short for "Retrieval-Augmented Generation".
Standard AI answers using only its pre-trained knowledge. Therefore, it cannot handle the latest information or internal company documents.
When using RAG, the AI searches for relevant documents before answering and then responds. It's like "taking a test while looking at a cheat sheet."
Google's NotebookLM is an easy-to-understand example; it answers based only on the files you provide. You can also check on the screen where the basis for the answer came from.
2. Multimodal AI
"Modal" refers to the type of information.
Conventional AI was primarily based on typing text to ask questions. Multimodal AI supports multiple formats, such as showing photos, speaking with your voice, or providing files.
Asking "What are the calories in this dish?" while showing a photo, transcribing recorded audio, or providing a PDF to be summarized. These ways of using it are already a reality.
ChatGPT, Gemini, and Claude are all multimodal-capable.
3. Microsoft Copilot
The name comes from the co-pilot of an airplane.
The person flying the plane is still the captain (=you). The philosophy behind this name is that the AI is just there to support you from the side.
Microsoft has integrated this AI directly into Word, Excel, Teams, and Outlook. The biggest feature is that you don't need to learn new tools, and the AI is right there in the screens you always use.
Automatic generation of meeting minutes, drafting email replies, and automating data analysis. These can now be done just by giving instructions in plain language.
4. Fine-tuning
This is a technique for additionally training existing AI with specialized knowledge or specific styles.
As they are, AI models can only provide generic answers. Fine-tuning is what allows you to teach them that "personal touch," such as your company's tone, industry-specific terminology, or specific formats.
The image of "teaching a new employee company rules and culture through training" is the closest analogy.
It is often confused with RAG, introduced in 1, but you should distinguish them by saying: use RAG if you want to reference the latest information, and use fine-tuning if you want to instill styles or specialized knowledge.
5. AGI (Artificial General Intelligence)
Short for "Artificial General Intelligence."
Current AI models have specific areas of expertise. AI that is good at writing text, AI that is good at image recognition. These are called specialized AI.
AGI is an entity that transcends those boundaries, referring to AI that can perform any intellectual task just like a human.
What is important is that AGI does not yet exist. Opinions among experts are also sharply divided regarding when it will be realized. While it is a dream technology, the "risk of it becoming uncontrollable" is also being seriously debated, making AI safety research an urgent priority worldwide.
Summary
✅ RAG (Retrieval-Augmented Generation)
AI that answers after searching through documents
✅ Multimodal AI
AI that can handle more than just text
✅ Microsoft Copilot
AI integrated into your usual tools
✅ Fine-tuning
Technique for additionally training AI with specialized knowledge
✅ AGI (Artificial General Intelligence)
AI that can perform any task (does not yet exist)
Thank you for reading. See you in the next article.