Ludovic Salenne: Mastering Generative AI Without Errors

Making the Most of Artificial Intelligence in Daily Life

What Are the Most Common Mistakes When Starting with Generative AI?

Ludovic Salenne, an artificial intelligence expert, identifies two major mistakes that beginners often make with generative AI. The first mistake is addressing an AI as one would a close friend. The interfaces of generative AIs, such as ChatGPT, are designed to establish immediate trust, which can lead users to forget they are interacting with a machine. This familiarity can result in disappointing outcomes if users do not master the best practices of prompt engineering. The results may then seem mediocre or even unusable, which can discourage users and lead them to abandon AI.

The second common mistake is getting distracted and searching for the ultimate AI. With new AI tools emerging daily, it becomes challenging to choose which one to use. Many fall into the myth of the supreme AI, supposedly capable of surpassing all others. However, this ultimate AI does not exist. Each AI has its strengths and weaknesses, and it is crucial to select tools based on one's goals and processes. The real mistake is testing all tools without taking the time to master them, leading to frustration and unsatisfactory results. Users may then feel lost and abandon AI until it is too late.

Salenne advises starting with a single tool and progressing with it, much like a musician who begins with one instrument to develop a solid foundation before exploring others.

How to Build Better Prompts?

To improve the quality of interactions with AIs, Salenne recommends creating your own prompts. He poses the classic question: would you prefer to be given fish or taught to fish? Two simple methods can help get started:

• RCT Method: This method involves structuring the prompt into three parts:

  • Define a role for the AI (for example: "you are an SEO expert..."),
  • Detail the context of the request,
  • Describe the task that the AI needs to perform.

• Prompt Improvement: Another approach is to ask the AI to improve one of your prompts. If the result is not satisfactory, explain why, and the AI can enhance it. Then, ask it to revisit your initial prompt and optimize it for the best outcome in the next use.

Should You Use Claude Instead of ChatGPT?

Salenne answers clearly: no, there is no reason to prefer Claude over ChatGPT. There is no ultimate AI. While Claude may be better for certain tasks like writing, it depends on mastering the specific prompts for that tool. Moreover, what is true today may not be true tomorrow. ChatGPT may outperform Claude in many aspects, and Gemini could be better than both for other tasks. Ultimately, all these AIs are overqualified for most tasks you wish to accomplish.

His advice is to choose the tool that appeals to you the most, test it, and integrate it into your processes. If the results do not meet your expectations, reflect, improve, and you will become an expert with that AI. Like an experienced pianist who can easily switch to guitar, you will have no trouble changing AIs to adopt the best one of its time.

What Are Your 5 Favorite AI Tools Right Now?

Ludovic Salenne shares his current favorite AI tools:

• Claude Cowork: an AI agent that works 24/7,
• NotebookLM: a free multitasking application for summarizing content, creating podcasts, making mind maps, slides, videos, infographics…
• n8n: for creating custom and automated AI agents,
• Notion AI: for planning, organizing, and producing,
• Suno AI: for creating music and entertaining your friends during gatherings!

However, he notes that these preferences can change rapidly, and he might recommend five different tools the very next day.

Understanding and Evaluating Artificial Intelligence

How to Know if a Response Generated by an AI is Reliable?

According to Salenne, the reliability of a response generated by an AI depends on the reliability of the initial request. If you write a precise and well-formulated prompt, there is little reason for the AI to be mistaken. However, he acknowledges that AI hallucinations exist and it is important to be aware of them.

He advises following two principles:

• Ask the AI to rely on reliable sources, share them with you, and not to invent anything.
• Use multiple AIs for the same request. Salenne applies this method when working on complex and sensitive topics, soliciting multiple AIs to compare and combine responses. This proves to be very effective, as each AI is configured according to the values of its creators and influenced by their training methods.

Do Social Media Really Influence the Responses of Generative AIs?

Salenne confirms that social media influences generative AIs, but to a certain extent. Large Language Models (LLMs) are trained on vast volumes of data and web content, including elements from social media. However, this influence manifests during the training of the AIs and not in real-time during a request.

AIs are well-configured to search for information when necessary, based on reliable sources. Websites with good authority are prioritized for AIs, while social media are clearly not. Moreover, most AIs do not have access to social media content during their production. Social media generally have their own AIs (Meta AI for Facebook and Instagram, Grok for X…) and thus close the doors to competitors.

Salenne compares this to the secret recipe of Coca-Cola: we know how it works, but we do not have the precise recipe.

AI Serving Digital Professionals

How to Conduct Sector Monitoring with AI? What Tools Are Recommended?

To conduct effective sector monitoring with AI, Salenne distinguishes between two approaches:

• Manual Approach: Use an AI at a determined pace, connect, and conduct your research with predefined prompts. An interesting tool for this is Perplexity, designed for research and very effective for monitoring.

• Automated Approach: This is clearly the ideal! You can regularly receive a comprehensive report according to your needs. For this, Salenne recommends using the no-code automation tool n8n with its "AI Agents" feature.