What AI Is Genuinely Bad At
Lesson 1.4 — What AI Is Genuinely Bad At
Understanding the limits makes you a better user
People who get frustrated with AI usually fall into one of two traps: they expect too much and get burned, or they dismiss it entirely and miss something genuinely useful. This lesson helps you avoid both.
Knowing where AI struggles doesn't make it less useful — it makes you more effective, because you know when to trust it and when to verify.
1. Making things up (hallucination)
This is the big one. AI can generate text that sounds completely confident and authoritative — and be completely wrong.
The technical term is hallucination: the AI produces plausible-sounding but false information. It might invent a fake research paper with a realistic-sounding author and journal. It might confidently give you the wrong phone number for a business. It might state a historical "fact" that never happened.
This happens because the AI is predicting what sounds right, not retrieving verified facts from a database. When it doesn't know something, it doesn't always say "I don't know" — sometimes it just fills in what seems statistically likely.
What to do: Always verify important facts from AI using reliable sources. Don't use AI-generated content for anything safety-critical without checking it.
2. Knowing what happened recently
Most AI models have a knowledge cutoff date — a point after which they weren't trained on new information. Ask ChatGPT about something that happened last month and it may not know, or it may give you outdated information as if it were current.
Some AI tools now have web browsing built in (like Perplexity, or ChatGPT with browsing enabled), which helps — but it's still worth checking when the AI last updated its knowledge.
What to do: For recent events, news, or rapidly changing topics, use an AI with live web search, or verify with current sources.
3. Maths and precise calculations
It sounds counterintuitive — computers are supposed to be great at maths, right? But language models are not calculators. They generate answers by predicting likely text, not by computing. Simple arithmetic is usually fine, but complex calculations can go wrong.
For example, a language model might confidently get a multi-step maths problem wrong because it's completing the pattern of a maths answer rather than actually running the calculation.
What to do: Use a calculator, spreadsheet, or code for precise maths. You can ask AI to set up a calculation or explain the formula, then do the actual number-crunching elsewhere.
4. Knowing what's happening right now in your specific life
AI has no memory between conversations (unless you've specifically set up memory features), no access to your calendar, your emails, or your files unless you share them. It doesn't know your context unless you tell it.
This is actually a good thing for privacy — but it means you need to give the AI relevant context when you start a conversation.
What to do: Think of AI like a brilliant consultant who just walked into the room. They're smart, but they need you to brief them on the situation before they can help.
5. Consistently reliable legal, medical, or financial advice
AI can be a useful starting point for understanding legal, medical, or financial topics — explaining concepts, summarising information, helping you prepare questions for a professional. But it shouldn't be your final source of truth for high-stakes decisions.
It can miss important nuances, get jurisdiction-specific details wrong, or give you general information that doesn't apply to your specific situation.
What to do: Use AI to understand and prepare, then consult a qualified professional for important decisions.
6. Common sense reasoning (sometimes)
AI can write a brilliant essay but struggle with a simple riddle. It can explain quantum physics but sometimes fail at "if John is taller than Mary, and Mary is taller than Pete, who is shortest?"
This is improving rapidly with newer models, but it's still a real limitation. AI reasoning can break down on tasks that require genuine step-by-step logic, especially when the problem is designed to be tricky.
What to do: For complex reasoning, ask the AI to "think step by step" (this genuinely helps) and check its logic manually for important decisions.
The bottom line
| AI is great at | AI struggles with |
|---|---|
| Writing and editing | Accurate facts (verify everything) |
| Explaining concepts | Recent events |
| Brainstorming ideas | Precise calculations |
| Summarising text | Your personal context |
| Translating languages | High-stakes professional advice |
| Creative tasks | Tricky logical puzzles |
Key takeaway
AI's biggest weakness is that it can be confidently wrong. The solution isn't to distrust it entirely — it's to verify important claims, understand what it's being used for, and treat it as a powerful assistant rather than an infallible authority.
Next up: Lesson 1.5 — AI Myths vs. Reality