Understanding Model Capabilities
What AI models can and cannot do, and how to choose the right model for your task.
Knowing an AI model's strengths and limitations prevents frustration and helps you craft prompts that play to the model's strengths. AI is not a general intelligence — it's a powerful pattern matcher with specific capabilities and blind spots.
- Summarization: Condensing long documents into key points
- Translation: Between languages, formats, or technical levels
- Rewriting: Changing tone, style, format while preserving meaning
- Brainstorming: Generating many ideas quickly across diverse angles
- Structured output: Converting unstructured text into JSON, tables, or other formats
- Code generation: Writing, explaining, and debugging code
- Analysis: Breaking down complex topics into components
- Precise math: LLMs sometimes make arithmetic errors, especially with large numbers
- Real-time information: Models have a knowledge cutoff date and can't browse the web (unless given tools)
- Counting and spatial reasoning: Counting letters in a word or items in a list can be unreliable
- Consistent memory: Long conversations may lose earlier context
- Factual accuracy: Models can hallucinate — stating false information confidently
- Original research: AI synthesizes existing knowledge but doesn't generate new discoveries
Different models have different strengths. Smaller, faster models are better for simple tasks. Larger, more expensive models handle complex reasoning better. Here's a practical framework for choosing:
- Simple tasks (rewriting, formatting, extraction): Use the fastest, cheapest model available. GPT-4o mini, Claude Haiku, Gemini Flash.
- Standard tasks (writing, analysis, code): Use mid-tier models. GPT-4o, Claude Sonnet, Gemini Pro.
- Complex reasoning (multi-step analysis, difficult code, nuanced writing): Use the most capable model. Claude Opus, GPT-4o with reasoning, Gemini Ultra.
- When in doubt: Start with a mid-tier model. Only upgrade if the output quality isn't sufficient.
Prompt Templates
Fact-Check Request
Gets the model to self-assess its confidence in factual claims.
I need you to analyze this claim: "[CLAIM]". Rate your confidence in its accuracy on a scale of 1-10. Explain what parts you're confident about and what parts you're uncertain about. If you're not sure, say so clearly — don't guess.
Task Difficulty Assessor
Helps you decide if and how to use AI for a specific task.
I want to use AI for this task: [DESCRIBE TASK]. Assess: Is this a good use case for AI? Rate it 1-10 on feasibility. What are the risks? What model tier (fast/standard/powerful) would you recommend? What should I verify manually?
Test Your Knowledge
Knowledge Check
1 / 2
What is "hallucination" in the context of AI?
Key Takeaways
- ✓AI excels at summarization, rewriting, brainstorming, and structured output
- ✓AI struggles with precise math, real-time info, counting, and factual accuracy
- ✓Hallucination is the most dangerous limitation — always verify critical claims
- ✓Match model size to task complexity: simple tasks don't need the most powerful model
- ✓A great prompt on a small model often beats a lazy prompt on a large model
Continue Learning
What Are LLMs
A plain-English explanation of large language models and why they behave the way they do.
Tokens & Context Windows
Learn about tokens, context windows, and why they determine what AI can and cannot do.
Temperature & Sampling
Understand temperature, top-p, and other settings that control how creative or deterministic AI outputs are.