AI Fundamentals· Lesson 7 of 10
- 1What Even Is AI? AI vs. ML vs. Deep Learning6 min
- 2How Models Learn: Training vs. Inference6 min
- 3What Is an LLM?8 min
- 4Tokens — the Currency of Everything7 min
- 5Why AI Output Is Probabilistic (and Why It Changes How You Build)7 min
- 6Hallucination — Why Models Make Things Up7 min
- 7What AI Is Genuinely Good and Bad At — a Capability Map7 min
- 8AI Systems vs. Traditional Software — the Mindset Shift7 min
- 9Where AI Fits in a Product (and Where It Shouldn't)7 min
- 10Responsible AI Basics for Builders7 min
What AI Is Genuinely Good and Bad At — a Capability Map
In one line · AI excels at language, patterns, and summarizing; it's weak at exact math, guaranteed facts, and long reasoning.
Knowing the limits before you build lets you pick the right tasks for AI and avoid handing it work it does poorly.
Think of it as a brilliant language employee who's shaky with numbers. Give it writing and understanding; leave sensitive math to ordinary code.
Summarizing a 10-page report = great. Summing 200 numbers exactly = use code, not a model.
Apply the map with three questions about any task: (1) Is it fundamentally a language task (understanding, summarizing, classifying, generating)? If yes, the model is in its zone. (2) Is an occasional error acceptable and easy to review? If a single error is catastrophic, you need a human or code alongside. (3) Does it demand 100% precision (math, dates, rule-matching)? If yes, that's ordinary code's job — or have the model call a tool for that part. The map shifts with every model generation, so re-check it rather than memorizing it.
🧪 Try it: list 5 real tasks from your own work (e.g., replying to an email, computing a discount, summarizing a meeting, reviewing a contract, triaging complaints). Classify each: model, ordinary code, or both together? Then actually try one 'model' task and see if your classification held.
Common mistakes: (1) assuming 'speaks well' means 'computes well' — confident style is no proof of accuracy; (2) rejecting AI entirely after one visible failure — the fix is usually changing the task, not scrapping the tool; (3) judging capabilities from a two-year-old article — the map moves fast.
Use the map at the start of any AI feature: deciding what to hand the model, what stays ordinary code, and where a human belongs in between.
Don't treat the map as eternal law: capabilities improve every release, and yesterday's weaknesses become acceptable. Test and measure on today's models instead of relying on stale impressions.