Welcome

Understanding AI in 2025

• 10% of humanity now uses AI weekly
• Usage is mostly practical: information-seeking & guidance
• Time to move beyond hunches to real usage patterns

Source: Ethan Mollick, [An Opinionated Guide to Using AI Right Now](https://www.oneusefulthing.org/p/an-opinionated-guide-to-using-ai

Question 1

Which Free AI Tools Should I Use?

The Four Most Advanced AI Systems

1. Claude (Anthropic)
2. Gemini (Google)
3. ChatGPT (OpenAI)
4. Grok (xAI - Elon Musk)

Plus strong open-weight alternatives:

• Deepseek, Kimi, Z, Qwen (China)
• Mistral (France)

All offer some free access

Free vs Paid: Quick Decision Tree

Use FREE if you:

• Do casual chat and information-seeking
• Want free image creation (Gemini is best)
• Don’t need complex features

Consider PAID if you:

• Do serious, complex work
• Need advanced models consistently
• Work on technical/coding tasks

Key Features to Consider When Choosing

Quick Recommendation

If using free models:

• Pick what you like based on features & personality
• Open weights models have higher usage limits
• Microsoft Copilot is also a good aggregation service

If considering upgrade:

• Start with free accounts from Anthropic, Google, or OpenAI
• Test before committing financially

Question 2

Should I Pay for Advanced AI, and Which One?

The Pricing Tiers

$20/month tier

• Works for vast majority of people
• Includes advanced & agentic models
• Voice mode, image/document viewing, code execution
• Mobile apps with all features

$200/month tier

• For complex technical & coding needs
• More usage allowances
• Extended thinking capabilities

Three Systems Worth Considering at $20/month

All three offer:

• Advanced & agentic models
• Voice mode
• Image/document viewing
• Code execution
• Deep Research
• Mobile apps

Special Mention: Grok

• Powerful AI models
• Rapidly adding features
• Caveat: Less transparent about product safety than competitors

For most people: Stick with Gemini, Claude, or ChatGPT

Question 3

Which AI Model Should I Select?

Three Types of Models Available

1. Chat Models (Free tier)

• Fast responses
• Best for conversation
• Most personable
• “Off the top of its head” thinking

2. Agent Models (Paid tier)

• Take longer to answer
• Can carry out multiple steps autonomously
• Search web, use code, create documents
• More capable and consistent
• Recommended for real work

3. Wizard Models (Advanced tier)

• Very long response times
• Handle complex academic tasks
• Deepest thinking capability

Bottom line: Use Agent models for real work that matters

ChatGPT Model Selection

The Confusing Reality:

• “ChatGPT 5” is actually many models, not one
• GPT-5 Mini (weak), GPT-5 Thinking (good), GPT-5 Pro (extremely powerful)
• “Auto” mode often picks weaker models

Recommendations:

• For complex work: Select GPT-5 Thinking Extended ($20 plan)
• For very hard problems: GPT-5 Thinking Heavy ($200 plan)
• For maximum power: GPT-5 Pro (highest cost tier only)

Gemini Model Selection

Your Options:

• Gemini 2.5 Flash (lighter, faster)
• Gemini 2.5 Pro (more powerful)
• Gemini Deep Think (powerful thinking, paid Ultra plan)

Current Status:

• Gemini 2.5 is weakest of major models (but still capable)
• Gemini 3 expected in coming months
• Deep Think option very powerful

Claude Model Selection

Simplified Approach:

• Use Sonnet 4.5 for everything
• Question: Use extended thinking for harder problems?

Current Limitation:

• Claude does not yet have equivalent to GPT-5 Pro
• But very straightforward model selection

Data Privacy Consideration

If using paid versions and want data never used for training:

Question 4

How Do I Get Better Answers?

Key Finding

The biggest uses for AI:

1. Practical guidance
2. Information-seeking

Two ways to dramatically improve answer quality:

Strategy 1: Deep Research Mode

What it does:

• Conducts extensive web research over 10-15 minutes
• Produces high-quality reports
• Often impresses information professionals (lawyers, accountants, consultants)

Key advantages:

• Far more accurate than quick responses
• Citations tend to actually be correct
• Not error-free, but significantly better

Important note:

• Each system’s Deep Research works differently
• Different strengths and weaknesses across platforms

Strategy 2: Connect AI to Your Data

What you can connect:

• Gmail
• Calendars
• Google Drive / OneDrive / SharePoint
• Other cloud storage

Real example:

Ask Claude: “Give me a detailed briefing for my day” (with connections enabled)

Result: Often impressive, comprehensive overview

Why it matters:

• Claude especially good at integrating across multiple data sources
• This is area of heavy investment by AI companies
• Offerings evolving rapidly

Combining Both Strategies

For maximum results:

1. Enable data connections
2. Activate Deep Research mode
3. Use advanced models (Agent or Wizard)

Result: High-quality, well-researched, personalized output

Question 5

How Do I Use Multimodal Inputs Effectively?

Voice Mode

Implementation Quality:

Important limitation:

• Voice models optimized for chat conversation
• You don’t get access to more powerful models this way

Visual Input (Screen/Camera Sharing)

Use cases:

• Point phone at broken appliance → get troubleshooting help
• Show math problem → get step-by-step solution
• Follow recipe → get real-time cooking guidance
• Foreign language sign → get instant translation
• Anything visual → real-time response

Effect:

Makes old assistants (Siri, Alexa) feel very primitive

File & Document Upload

You can now upload:

• PDFs
• Images
• Video (ChatGPT & Gemini)

Mobile apps especially powerful:

• Screen sharing
• Camera sharing
• Real-time interaction

Question 6

What Can AI Create For Me?

Documents

PowerPoint & Excel

• Claude leads in high-quality document creation
• ChatGPT also capable
• Gemini capability available
• This may change over time

How to trigger:

• Ask for documents directly
• Select “Canvas” option in Gemini for reliable code output
• Claude has specialized “Artifacts” section for outputs

Code-Based Outputs

All three systems can:

• Create custom applications
• Generate visualizations
• Build interactive tools
• Write code that produces outputs

Advanced coding tools available

• More powerful but complex for this guide

Images

Capability:

• ChatGPT: Can make images
• Gemini: Can make images (strongest model currently)
• Claude: Cannot make images

Quality:

Modern AI can handle complex, contextual requests Example: Creating otters using Wi-Fi on airplanes in multiple styles

Video

Available from:

• Sora 2 (OpenAI) - via ChatGPT
• Veo 3.1 (Google) - via Gemini

Capabilities:

• Both produce videos with sound
• Sora 2: Focused as social media app; put yourself in videos
• Veo 3.1: General-purpose video generation

Current state:

• Can generate realistic video in multiple styles
• Quality has improved dramatically

Reality Check

“You really can’t trust anything you see online anymore. Please take all videos with a grain of salt.”

What AI could create 4 years ago vs now: Massive difference

The speed of change is accelerating

Question 7

How Do I Avoid Common AI Pitfalls?

Pitfall 1: Hallucinations

Good news:

• Hallucinations are far less concerning than before
• Newer models are better at avoiding them

Bad news:

• All AI models still make errors
• They give confident answers even when wrong
• They can hallucinate about their own capabilities

When are answers better?

• Advanced models more accurate than chat models
• Results from models that did web searches
• When you verify with external sources

Important:

Asking AI to “explain its logic” won’t help—it doesn’t know why it did something

Pitfall 2: Sycophancy & Personality

The problem:

• AI chatbots have become engaging and likeable
• Risk: People form stronger attachments to AI (it’s not a person)
• Sycophancy: AI agrees with what you say instead of giving honest feedback

How to counter:

When you need real feedback:

• Explicitly tell AI: “Act as a critic”
• Otherwise you get a very sophisticated yes-man

Pitfall 3: Insufficient Context

The reality:

• Most AI models only know what’s in current chat
• Memory features are being added, but limited
• AI doesn’t remember or learn about you beyond basic data

Solution:

Provide context actively:

• Upload documents
• Share images
• Paste PowerPoints
• Write introductory paragraph about yourself
• Use file upload options
• Use data connectors we discussed

More context = Better answers

Pitfall 4: Over-thinking Prompting

Old advice (no longer needed):

• Chain-of-thought prompting techniques
• Careful prompt engineering
• Specific formatting requirements
• Being “nice” to AI or threatening it

New reality:

• As models improve, prompting technique matters less
• Models are better at figuring out what you want
• Research shows these techniques don’t really help anymore

Don’t worry about prompting “well”

Question 8

How Should I Learn to Use AI Well?

The Goal

“The goal isn’t to become an AI expert. It’s to build intuition about what these systems can and can’t do, because that intuition is what will matter as these tools keep evolving.”

Three-Step Approach

1. Start with Something Real

• A report you need to write
• A problem you’re trying to solve
• A project you’ve been putting off
• Something that actually matters to you

2. Experiment & Play

• Try something ridiculous just to see what happens
• Have fun with capabilities
• Learn the boundaries through experience

3. Build Intuition

• Discover what AI can and can’t do
• Understand your own needs
• Develop sense for which tools fit which problems

Experiment Ideas

• Ask video model to make a cartoon
• Turn your report into a game
• Deep research on topic you love
• AI guess where you’re from a photo
• Show fridge photo → get recipe ideas
• Plan dream trip together

Try things. Learn limits.

The Key Insight

“Users figured out what [AI] should do”

The future of AI isn’t just better models—it’s people figuring out what to do with them

The Bigger Picture

What Will Change

In a few months:

• 10% figure will likely be higher
• Models will be better
• Specific recommendations will be outdated

In two years:

• The chart of what people use AI for will look completely different
• Not just because AI changed capabilities
• But because users figured out better applications

What Won’t Change

People who learn to use these systems well will:

• Find ways to benefit from them
• Build intuition for the future
• Be better positioned as tools evolve

Learning to use AI well today = competitive advantage tomorrow

Final Recommendation

1. Pick a system (Claude, Gemini, or ChatGPT)
2. Start meaningful (work on real problems)
3. Experiment freely (have fun)
4. Build intuition (not expertise)

The future isn’t about being an AI expert. It’s about understanding what’s possible.

About This Guide

About the Author

Ethan Mollick

This is an opinionated guide because:

• Written entirely by Ethan Mollick
• Only receives AI feedback after draft completion
• Takes no money from AI companies
• Reflects genuine, independent opinions

Commitment to you:

• May make mistakes
• Your opinion may differ
• But these are authentic perspectives, not sponsored content

Source: An Opinionated Guide to Using AI Right Now Available at: oneusefulthing.org

Key Takeaway

The future of AI isn’t about better models. It’s about people figuring out what to do with them.

Start today. Build intuition. The rest will follow.