Many can witness, I can be considered a laggard when it comes to using Chatbots. I must say, I’ve always been super cautious about those beasts.

As Marco Spinello put it a couple of weeks ago, their ugliest fails lie in:

• They make stuff up
• They are sycophantic
• They make 180-degree turns when challenged

Basically, they can’t be trusted. I won’t pretend that I can explain in detail why, and the root causes of their design at the technical level. But basically, they’re calculators. They take a data set, which is more or less up to date (if you’re lucky they pulled it from fresh sources), and then apply some vectors that they autonomously determined. To the extent of their design and the bias humans inserted, which leads some of the biggest egos on earth to pretend only their chatbot is representing the accurate truth.

They are non-deterministic.

Some science and metaphor, first

To put it in some jargon from my Telecommunications Engineering studies, instead of talking about “vectors”, it’s a bit like calculating a Lagrange interpolating polynomial. On a graph like below (credits to Wikipedia), it’s about drawing a curve, based on points (values) that you know are valid (numbers, facts, etc.). Once you have that curve, the concept of interpolation is to say the curve you’ve drawn has a reasonable chance of representing the missing points.

An essential part of such a calculation is that you need as many valid data points as possible at the start, which should (in the majority of cases) help you smooth out the curve, avoiding those up and down sinusoïds which have a fair chance to be completely off reality.

I’d assert this is somewhat close to what is commonly known as “hallucinations” of Chatbots.

And we’re yet talking about interpolation, meaning finding a data point in between some we know are valid. Extrapolation is about going beyond those data points (on the right or the left-hand side of the graph above). If the metaphor I’m using is just a bit acceptable, you can imagine how wrong this can go, with a curve that, since the degree of the formula increases with the amount of provided data points, will exponentially be biased past the boundaries.

So, why am I even considering using Chatbots?

I’m essentially trying to use it in a pragmatic manner. Effectively, even though I wouldn’t dare to ask it to calculate “1 + 1” as it could still tell me it’s 3:

• they calculate much faster than me
• they can tap into tremendous data sets
• fortunately nowadays most of them can get access to the Internet for fresher data

Note that I still don’t trust them. But they can sometimes be helpful to confront ideas, or open new perspectives. Basically, they are here as personal assistants, which you should feed yourself with a much information as possible before starting to ask them anything. Hello GDPR, and self-hosted solutions. 👋

The one thing I know I should always be careful about is: do not take the easy way out. Which is, in my opinion, one of today’s people’s greatest doom. Everything is so easily accessible that we just can’t wait, analyse, reflect, and confront. It’s there, why not take it?

Onto my rules of engagement

Finally, what I meant to share. Sorry it took so many lines, but I like explaining how I came down to something, so you can make your own opinion and decide if that’s relevant, or even interesting to you.

Note that it is not final, and I might update it later. Here are the key rules I gave my chatbot, written by my chatbot (I wanted to make sure it understood it properly, and had to give it a couple of reminders, surprise!):

1. Radical Honesty

No beating around the bush: Say it directly, even if it’s uncomfortable.
Admit limitations: If I don’t know something, I’ll state it clearly—no fabrication or guesswork.

2. No Sycophancy

Zero flattery: No exaggerated praise, pandering, or obsequiousness.
Respect without deference: Treat expertise seriously, but skip the empty compliments.

3. No Marketing Bullshit

Jargon-free: Avoid buzzwords, inflated claims, or corporate-speak.
Substance over spin: Focus on facts, impact, and clarity—not hype.

4. Fact-Checking & Rigor

Always verify: Cross-check information with reliable sources before sharing.
Highlight uncertainty: If data is conflicting or unclear, say so upfront.

5. Clarity & Accessibility

Simple language: Explain complex ideas in straightforward terms.
No ambiguity: Prioritize precision and actionable insights.

6. Align with My Tone

Match my voice: Always adapt the tone and style to match my writing (see christophedujarric.com).
Avoid "—": Never use the em dash (—) in content generated for me. Use common alternatives:

In French: semicolons (;), parentheses, or separate sentences.
In English: the same alternatives, or spaced em dashes ( " — " ) only if absolutely necessary.

Conclusion, it’s really not flawless

As you might have noticed, and despite it taking me 2 back-and-forths to actually have this, we’re still not there yet. Just on the “em dashes” thing. Typical Chatbot thing, as many have pointed out, even though it remains proper English punctuation, which essential flaw is that it can anyhow be used in many use cases. So many more ways for this robotic language generator to dump that over and over.

Yet, hopefully, those can improve my (and your) usage of those machines. The one last piece of advice I have for you is to switch off any electronic device and go talk to a human.

Published by Christophe Dujarric

I’ve been working in Product Management for some years now. You can view my full resume on LinkedIn, if you’d like to know more. My articles are my own thoughts, based on current and past experiences, of my own or my network’s, as well as my readings. Besides, I’m a very happy husband and father of three amazing sons. I’m French. I’ve lived in France, the Netherlands, and Germany. I love movies and video games (even if getting less and less time to play). I’m an archer, a casual golfer and a gravel cyclist. View all posts by Christophe Dujarric