"Think of AI models like coworkers: Some are chatty and creative, some are detail-obsessed, and some are pure logic. Use them wisely."

Every week, a new model drops: o1, DeepSeek R1, Gemini 2.0 Flash, o3 Mini… What does it all mean?

IT…IS... CONFUSING!!! Right?

I will share how I make sense of it all.

I basically look at 3 different types of AI models and in this post I will help you understand:

1. Which model to use?

2. How to get the most of the model?

"You wouldn’t ask a poet to do your taxes or an accountant to write wedding vows—so why expect one AI model to do everything?"

The Three Types of AI Models (I mostly use)

NOTE: I hardly use AI for image generation, so I didn’t include any image AI here.

Depending on the task, I mostly use three types of models.

NOTE: only the Chat (Generative) AI is a traditional GPT —> Large Language Model.

AI models have different STRENGTHS, and knowing who to ask for help (and who to avoid for certain tasks) will save you time and frustration.

Choosing the Right Model for Your Use Case:

• Prioritise speed and cost? → Chat (Generative) AI models are faster and more affordable.

• Need to execute well-defined tasks? → Chat (Generative) AI models handle them efficiently.

• Require accuracy and reliability? → o-series reasoning models excel in decision-making.

• Solving complex problems? → o-series reasoning models handle ambiguity and multi-step reasoning.

If your tasks are straightforward and speed matters most, GPT models are the best fit. If accuracy and complexity are key, choose o-series models.My 

AI Coworkers (Let’s have some fun)

I like to think of the different AI models as coworkers. Each of these models has a different personality and skillset—some are friendly conversationalists, others are strict logicians, and some are meticulous researchers.

Let’s meet the team.

🗣️ The Chat (Generative) AI Models

The creative conversationalist—brilliant at brainstorming, storytelling, and engaging discussions but sometimes a little too confident.

These are your GPT models.

ChatGPT 4.0 – "The Wise Mentor Friend"

Best for: Brainstorming, content creation, casual summarisation, engaging explanations.

Weaknesses: Overconfident (hallucinates facts), over-explains, not ideal for structured tasks.

Can be loooooooooong-winded.

How to Use:

• Works best with conversational follow-ups—refine your request through back-and-forth dialogue.

• Great for open-ended and creative prompts ("Help me write a fun blog post about AI!").

• If it rambles, say "Make this shorter" or "Summarise in one sentence."

• Learn Prompt Engineering (how to write great prompts).

🔎 Example Prompt:
"Explain quantum computing like I’m 10 years old, but make it funny."

I use ChatGPT (I have the pro plan), but switch to Gemini and Claude to test and explore (They are also great LLMs)

Chat (Generative) AI Model Comparisons:

 Claude Sonnet 3.5 – More thoughtful and careful with logic than ChatGPT but a bit more reserved.

Gemini 1.5 Pro – More factually accurate and better at summarising but less engaging for creative brainstorming. (Gemini is from Google)

👉 If you want balanced conversation + reliability, Claude is great.
👉 If you need deeper research, Gemini is better.

But sometimes, you need less talk and more action. Enter the reasoning AI models—the no-nonsense, detail-obsessed perfectionists of the AI office.

🧠The Reasoning AI Models

The no-nonsense problem solvers—great for structured thinking, coding, and math, but not the ones you chat with casually.

OpenAI has a Guide to using Reasoning Models.

Key Differences Between Reasoning and Non-Reasoning GPT Models:

• Reasoning models (e.g., o1 series) are optimized for complex, multi-step problem-solving.

• They use "chain-of-thought" reasoning, generating intermediate steps before arriving at a final answer.

• Best for advanced tasks like math, science, and complex coding challenges.

• Non-reasoning models (e.g., GPT-4o) focus on speed and versatility, handling general-purpose tasks more efficiently.

When to Use the Reasoning Models:

• Navigate Ambiguous Tasks: Excels in complex, multi-step problems like math, science, and coding.

• Prioritise Accuracy Over Speed: Best when precision and reliability are more important than quick responses.

• Tackle Deep Analytical Work: Ideal for structured reasoning and in-depth analysis.

How to Prompt Reasoning Models Effectively:

• Encourage step-by-step solutions → e.g., "Think step by step before answering."

• Specify the desired format → e.g., "Provide an explanation followed by a final answer."

• Provide relevant background information to improve context understanding.

• Use clear and specific language to minimize ambiguity and guide the model effectively.

These models work best with one-shot prompts. Imagine giving them a detailed brief instead of having a conversation.

One-shot prompting: You give the AI model a single example of the desired input and output to guide its response to subsequent prompts.

NOTE: This is one of the ways the reasoning models differ from an LLM (chat model) like chatGPT where you get better results when you are conversational.

OpenAI o1 – "The A-Type Overachiever"

The o1 model is included in the paid plan of ChatGPT.

 Best for: Structured reports, data organisation, documentation, step-by-step problem-solving.

Weaknesses: Bad at casual chat, requires highly detailed prompts.

How to Use:

• Works best for structured output.

• Rework your prompt if needed—it won’t refine answers iteratively.

• Effective Prompting is Crucial: Instead of short prompts, provide clear goals, formatting details, and context.

• Single-Shot vs. Iterative Interaction: Unlike ChatGPT, o1 aims to get things right in one go, so spend more time crafting your prompt upfront.

The skills and prompting techniques you use for ChatGPT 4.0 is different than what is effective for the o1 models...

Here is an example of the reasoning the o1 model does:

Here is an example of a great prompt for o1. The user wanted o1 to give an in-depth analysis of hiking trails in SF:

💰 o1 Pro – The High-Speed Executor

Need complex calculations, structured problem-solving, fast execution, and large-scale data processing? o1 Pro is your guy—but it comes at a cost: $200/month.

Comparison: DeepSeek R1

DeepSeek R1 – Another strong reasoning model, better at logic-based tasks than chat-based models.

⚠️ Heads-up: DeepSeek R1 may be banned in the US soon, and their privacy policy has changed—they now own all data processed through their model.

The Research AI Models

The deep-thinking, fact-checking specialist—prioritises accuracy over speed or creativity.

📖 Gemini Deep Research – "The Academic Scholar"

This is one of the perks of paying for a Gemini subscription. If you need facts, this is where you go.

🔍 How it works:
1️⃣ Develops a research plan (you approve it).
2️⃣ Searches the internet for for information (can take time, so be patient).
3️⃣ Writes an in-depth report with citations.

Best for: In-depth research, fact-checking, reports with sources.

Weaknesses: Slower than other models, not great for casual conversation.

How to Use:

• Best for research-heavy prompts ("Summarize the latest climate change studies with sources.").

• Ask for citations if needed.

• Expect slower but more factually accurate responses.

🔎 Example Prompt:
"Compare recent studies on the impact of AI in healthcare. Include key findings and sources."

Choosing the Right Tool for the Job

To maximise your AI team, consider:

• Task Complexity: Need quick ideas? ChatGPT. Need deep analysis? o1.

• Research Depth: Quick fact-checking? o3 Mini. In-depth research? Gemini Deep Research.

• Desired Output Format: Creative text? Chat models. Structured reports? Reasoning models.

Go and experiment. Get a feel for how each model works. 

Final Thoughts: Who Do You Turn to?

The good news for entrepreneurs is that you now have a whole host of coworkers. Do not be anti-social. Get to know them, especially what they are good at.

Same with AI—pick the right office buddy, and you’ll avoid frustration (and questionable results).

• Need ideas? Grab coffee with GPT-4 Turbo—they’ll talk your ear off, but you’ll get great insights.

• Need a structured report? Call o1—but expect zero fluff.

• Need fast data crunching? o1 Pro—it’s a machine (literally).

• Need a quick fact-check? o3 Mini—it’s like Google but faster.

• Need an academic deep dive? Gemini Deep Research—it’s basically writing a dissertation.

Moral of the story? Treat AI like a specialist coworker: use the right model for the right job.

And if all else fails… experiment and see what happens. 🚀

If you forget everything else, remember this…