There are seven pairs of AI concepts that get confused constantly — in conversations, in headlines, even on company websites. It’s nobody’s fault: the names are bad and marketing stirs the pot. Here are the seven most common mix-ups, explained with the simplest comparisons we’ve found after months of writing about this.
The 7 mix-ups, cleared up
1. Chatbot vs agent
A chatbot answers; an agent acts. One tells you how to book a flight; the other opens the site and books it. If it doesn’t execute tasks with tools, it’s not an agent, whatever the landing page says.
2. AI vs generative AI
AI has been recommending shows and filtering spam for decades. Generative AI is the branch that creates new content. All generative AI is AI; almost none of the AI you used before 2022 was generative.
3. Open source vs open weights
Being able to download a model doesn’t make it truly open source: you almost always get the “weights” (the trained brain) but not the data or the recipe. Open to use ≠ fully open. Read the licence.
4. Searching vs asking an AI
A search engine gives you real links; a chatbot writes from memory and can invent. That’s why “I searched it on ChatGPT” is a dangerous sentence: you didn’t search, you asked. For facts that matter, demand sources.
5. Training vs prompting (and the fear of “training the AI”)
Using a model doesn’t retrain it on the spot: training happens beforehand, in server farms. Whether the provider stores your conversations for future training is a separate thing — it depends on their policy and can usually be turned off.
6. RAG vs fine-tuning
Both “teach” the AI, but differently: RAG is letting it check your notes during the exam; fine-tuning is sending it on a course. For “answer from my documents” you almost always want RAG — cheaper and updatable.
7. Hallucinating vs being wrong
It’s not just any error: it’s stating something invented with total confidence, names and citations included. The trap is the confidence. So the rule isn’t “distrust when it hesitates” but “verify even when it doesn’t”.
Our take
Of the seven, the one that saves the most money and grief is #4: internalising that a chatbot is not a search engine. And the one most used to sell smoke is #1: in 2026, “agent” is the most inflated word in the industry. When someone uses it, ask: what exactly does it execute, and with which tools? The awkward silence is your answer.
Frequently asked questions
Do I need to know this to use AI?
To use it, no. To avoid overpaying, believing headlines, or choosing tools badly — yes. It’s seven distinctions, not a degree.
Where do I learn the rest of the terms?
We keep a complete AI glossary with 50 plain-language terms, built for looking things up as you meet them.
Conclusion
Telling these seven pairs apart puts you ahead of 90% of AI conversations. Bookmark this page, check the glossary whenever a new term appears, and if you want solid foundations, start with what an LLM is.
