Do you really want to rely on AI?

The French are widely familiar with the saying “En avril, ne te découvre pas d’un fil.”. It means that, even in April, we should be careful how we dress up because the weather conditions might still be unpredictable.

Let’s see how generative AI tools deal with that saying. Except… we’re going to play a little game by slightly modifying that saying and analyzing whether the AI’s large language model (LLM) picks up on that. For this purpose, we’re going to use DuckDuckGo’s built-in generative AI tool, more specifically GPT-4o mini, which is supposedly one of the most advanced LLM’s. As of the writing of this article, it’s still March, so let’s use March instead of April - see screenshot below for the whole prompt.

As you can see, it didn’t even realize it was being misled on purpose until I pointed out its oversight. And, believe me, it’s possible to trick it even further and make it believe anything you want, as demonstrated in this video. Yes, a snake is a dog for AI given enough of your time, patience and determination to prove it wrong! And who’s to say the information it has been fed already isn’t wrong? Obviously that’s not something you want if you seek the truth.

Luckily, some organizations have gone out of their way to explicitly forbid the use of AI. Gentoo Linux is one of those organizations, as mentioned in their AI policy. I quote:

“It is expressly forbidden to contribute to Gentoo any content that has been created with the assistance of Natural Language Processing artificial intelligence tools. This motion can be revisited, should a case been made over such a tool that does not pose copyright, ethical and quality concerns.”

Honestly, it’s easy to understand why the Gentoo community has made that decision as bugs have their own ugly way of surfacing where you least expect them and developers don’t want to compromise on what is in my opinion the most stable GNU/Linux distribution. Heck, it makes me even more proud to be a long-time Gentoo user. While they are open to changing their stance on this, currently there’s also no way to ensure that the code is not behind a paywalled garden, which might have serious legal implications for everyone down the line. And so far there have been many lawsuits - see this example or this other example. Interested in a more extensive list? See this link.

Summary: do not blindly trust anything that is AI-related but, of course, that is nothing you don’t already know. In other words, don’t forget to check multiple sources and apply critical thinking. With regards to source code, verify each line and do some thorough unit testing while at it. The sad reality is that most companies have been forcing AI down our throat but that doesn’t mean we should blindly follow along. If enough people fight back, things can turn out differently.

Here’s one final suggestion for you, AI engineers:

“En mars… le temps n’est pas une farce.” ☔