{"id":12440,"date":"2025-08-04T09:07:00","date_gmt":"2025-08-04T13:07:00","guid":{"rendered":"https:\/\/thirdwunder.com\/?p=12440"},"modified":"2025-08-06T15:48:55","modified_gmt":"2025-08-06T19:48:55","slug":"guide-de-lia-pour-les-neophytes","status":"publish","type":"post","link":"https:\/\/thirdwunder.com\/fr\/blog\/guide-de-lia-pour-les-neophytes\/","title":{"rendered":"Guide de l\u2019IA pour les n\u00e9ophytes"},"content":{"rendered":"\n

We\u2019ve all seen it before, generic and obviously AI generated content that starts with \u201cin a fast-paced digital landscape,\u201d and is full of all those words AI loves to overuse and abuse. It\u2019s led to constant callouts and gotcha moments in comment sections all over the web. But the biggest reason so much AI content sounds the same is because most people are still struggling with how to effectively use and prompt generative AI.<\/p>\n\n

So, I\u2019m here today to give you the non-tech-bro lowdown on how to get AI to actually do what you want it to\u2014no technical know-how required.<\/p>\n\n

What you need to understand about generative AI<\/h2>\n\n

We at Third Wunder have been doing a lot of experimenting with generative AI, trying to see how we can adapt and integrate this new technology into our workflows<\/a>. This has involved a lot of trial and error and not a small amount of frustration. But the biggest takeaway from all of it is this: AI is both very smart and very dumb.<\/p>\n\n

Let me explain. AI is able to compute what would take you or I several lifetimes to get through in a matter of seconds. But at the same time, it doesn\u2019t have our instincts or ability to make cognitive leaps. What that means is AI does not do nuance and it takes everything literally. So you need to be painfully specific and detailed in how you ask it to do things.<\/p>\n\n

\"\"\/<\/figure>\n\n

Sort of like the children\u2019s book character, Amelia Bedelia (pictured above), if you don\u2019t give AI the right context or leave it to its own devices, you\u2019re not going to be happy with its output.<\/p>\n\n

So, how do you get generative AI to do what you want it to?<\/h2>\n\n

By now, you\u2019ve probably heard a lot about prompt engineering. When generative AI first hit the scene, a lot of hay was made out of the idea that prompt engineering was going to become its own discipline. This, like many things in tech, soon fell to the wayside as newer, more sophisticated models have come out. That hasn\u2019t stopped people from talking about it.\u00a0<\/p>\n\n

One of the more persistent schools of thought from the early days of AI is the notion that you have to create the perfect prompt all in one go, almost as if you only get one chance to get it right. That couldn\u2019t be further from the truth. In fact, the more you ask AI to do in one go, the less likely you are to get something you\u2019re satisfied with. Which is why the most effective way to prompt AI is in the form of a conversation.\u00a0<\/p>\n\n

Think of it like briefing a freelancer. You want to be clear about the type of work you are asking for, as well as giving the AI pertinent information like your goals, intended results, and target audience. If it\u2019s content to be created, give it the necessary parameters like length, tone, and style, as well as examples of similar work. The more context the better.<\/p>\n\n

After that it\u2019s all about refining. Having a structured initial prompt is good, but the real work is in analyzing the AI\u2019s output and asking for refinements. It\u2019s a lot easier to articulate what exactly you\u2019re looking for after identifying everything you don\u2019t like about the draft in front of you.\u00a0<\/p>\n\n

What prompting AI looks like in practice<\/h3>\n\n

As I\u2019ve mentioned, prompting is more of an iterative process where you are in dialogue with the agent you\u2019re using. A lot of my own conversations go something like this:<\/p>\n\n

I usually start off with something along the lines of \u201cI\u2019ve been tasked with creating [content type] for [platform]. This piece is on [subject] from [angle\/perspective] and is targeted at [audience segment]. Can you help me create an outline?<\/em><\/p>\n\n

By asking the agent to write an outline or brief, it gives me a chance to see the direction and editorial choices the AI makes and it\u2019s easier at this stage to fine tune it, if the AI is missing key points or you want to go in a different direction with the piece. Once I\u2019m happy with the outline, I get the agent to start drafting the piece itself.<\/p>\n\n

Once you\u2019ve got a working draft, the conversation becomes more like a feedback or review session, where you can identify what you don\u2019t like and ask the agent to make changes. Telling it things like \u201ccan you add more meat?\u201d <\/em>or \u201cmake it sound more human and less boardroom,\u201d<\/em> go a long way towards getting output you\u2019re satisfied with.<\/p>\n\n

When in doubt, ask AI<\/h2>\n\n

Another approach is to ask AI to help<\/a> you create a prompt. If anyone is an expert on what AI can and cannot do, it\u2019s AI. That is to say, if you\u2019re not successful in getting the kind of output you need, the best solution is to ask AI why your prompt isn\u2019t working. This works for two reasons. One, the AI model you\u2019re using will identify any limitations or obstacles it\u2019s bumping up against (like your ask exceeding its memory limit). And two, this frees you up to more easily articulate your big-picture needs without getting bogged down with all the minute details.<\/p>\n\n

This is also an excellent exercise if you\u2019re not entirely clear on what it is you need. It\u2019s sort of like playing 20 questions, where at the end you have a clear set of directives<\/a> for what needs doing and how to get it done.<\/p>\n\n

What about all those pesky \u201cAIisms\u201d?<\/h2>\n \n