AI Tools: How to Choose?

Before Comparing Tools, Work on Your Prompts

AI tools are often judged on their raw performance, but one essential element is overlooked: the quality of the response depends as much on the prompt as on the model itself.

The Common Mistake

When faced with a disappointing response, many people get frustrated: “This AI is useless, it doesn’t understand anything.” Yet, a more detailed prompt is often all it takes to change everything. It makes sense: every generated word depends on the context. The more precise the context, the higher the probability of finding the right next word. Experienced users spend more time refining their request than reading the response—this is my case ����.

The Basics of a Good Prompt

• Clarify your intent: what you want to achieve and for what purpose.
• Assign a role to the AI: expert, teacher, journalist, popularizer…
• Specify the expected format: paragraph, list, table, structured outline.

Example

• Explain photosynthesis in 150 words, at a middle school level, with two concrete examples.
• Adopt the perspective of a life sciences teacher.
• Present the answer as a numbered list.

N.B.: The rise of AI reveals an irony: the better machines speak, the more crucial it becomes to speak better than them—or at least more accurately. Language is not a detail; it is the interface between our thoughts and their intelligence.

Evolution of AI Users (Estimates and Projections)

Relative to the 8 billion inhabitants of the planet, the number of AI users remains tiny, a sign that the revolution is only just beginning. In 2025, the ITU estimates that nearly three-quarters of humanity is connected, leaving 2.2 billion people offline, hence the surprisingly low number of users in 2025.

LLMs Are the Swiss Army Knives of AI

Large Language Models (LLMs) are the most well-known to the general public.
The capabilities of current AIs are so vast and their uses so diverse that no summary can list all the advantages and disadvantages; each tool opens up possibilities too broad to be confined to a simple table.
Here is a table based on my experience.

Common Limitations of All These Tools

Despite their feats, current AIs share structural weaknesses:

• Lack of genuine understanding – they manipulate probabilities.
• Bias and hallucinations – they invent facts and reproduce stereotypes present in the training data.
• Fragility to edge cases – a minor change in context, vocabulary, or word order can lead to inconsistent, contradictory, or incorrect responses.
• No physical grounding in the real world – unlike living beings, AIs do not interact with their environment. Their representation of the world mainly comes from data already produced by humans.
• Difficulty with long reasoning – current systems fail when it comes to following complex causal chains, maintaining consistency over long sequences, or reasoning outside frequently observed cases.
• Dependence on historical data – models build their responses based on statistical distributions learned from historical data. They extend observed trends rather than making true predictions.
• The "goldfish effect" – all AIs adapt to the user's style but retain nothing permanently. They adjust their responses to the immediate context. You often have to remind them of your rules, which can be frustrating. ����

Comparison of Free Quotas by LLM

Not all free tools are equal in terms of usage limits.

Key Takeaways

• Refine your prompts.
• To stay truly productive, use multiple AIs: you'll never be blocked by a restrictive quota.
• Don’t hesitate to submit the same prompt to them, analyze their differences, or even contradict them to refine the answer.
• And above all, keep a critical mind—they are just stochastic parrots.