It's been a year since I've been using ChatGPT daily, both professionally for recruitment consulting-related topics and personally for a variety of things.
I'm sharing my assessment of what I've managed to do with ChatGPT, and what I haven't been able to do.
What is ChatGPT?
It's not a magic wand; it's an Artificial Intelligence tool that bases its responses on what it has learned to answer your questions.
To simplify greatly, imagine that ChatGPT has read a lot (more than you and me), and depending on what you ask it, it will try to give you an answer based on everything it has read and the probability of what its answer should be.
If you want more details, there are technical articles that explain its functioning much better than I can.
In fact, I'm sure even ChatGPT would explain it better, let's ask it:
OK, thank you.
How often and for what purposes?
In a year, it appears I've used ChatGPT on at least 223 out of 365 days, which is 61% of the days. In fact, I say "at least" because the only information I have to judge are the dates of the conversations I've had with the tool.
However, I've used some conversations over several days, but ChatGPT only indicates the date of the last conversation update, so my use is probably higher than this figure.
40% of my conversations are professional,
60% are personal.
However, most of my conversations with ChatGPT are quite long, as they are workflows; I do research, I try things.
On the other hand, in a personal context, I've almost replaced Google with ChatGPT in most cases, because my personal use is mainly based on information search, and I save a lot of time in my searches this way.
The conversations are generally quite short, with few back-and-forths (except for some conversations that I continue as a personal assistant).
So, one could say that my professional use is more about production;
While my personal use is more about consumption;
Example: I've cooked, I have leftover egg whites, I ask ChatGPT "what can I do with 8 egg whites?". It then suggests a list of recipes; I choose one that I like, and I tell it "show me a recipe for sweet financiers that I can't mess up". ChatGPT then gives me precise advice so that I don't make a mistake in my recipe. I didn't need to search, compare recipes, and decide. I also didn't have to deal with various advertisements, pop-ups, cookie banners, etc.
I'm in a logic of pure information consumption, and I have little risk of not challenging the information given by ChatGPT: this info is just for me, and the only risk I run is messing up my financiers recipe (which could have happened without ChatGPT). So, I have everything to gain by not wasting time doing the info search myself and asking ChatGPT to do it for me instead.
What relevance for what efficiency?
A tool can only be as good as how it's used, and this is even more true for conversational AI.
We use queries, or "prompts," to ask ChatGPT to perform a task.
The better the task, its context, and the expected result are described, the better the result.
This is even truer for AI image generation tools, which, when pushed a bit, require not only specifying what one wants to see but also what one does not want to see (hey, it's good training for making a job description, don't you think?).
According to the uses described above, I invest time and effort into my prompts proportional to my need and stakes.
In other words, I speak to ChatGPT with a lot of preparation and care when I use it for work, and I converse with it more directly and lightly when my request is personal.
So, I would have 2 criteria in mind to judge the relevance and efficiency of my experiences and use with ChatGPT: the necessary effort (ROI) which is quantitative and the satisfaction of my request, which is qualitative and binary (it's satisfied, or it's not satisfied).
Why a binary threshold?
If I use a tool, it's not to do worse than what I would do alone.
I guess it's a matter of perspective, but I've never wanted to have nor offer a degraded experience in exchange for my free time. It's like with my clients: either I give my all, or I redirect to someone more effective than me if it's not my expertise.
What's the interest in having long conversations?
By "long conversation," we mean continuing the same conversation that has already started and continuing it over several days.
We're getting closer to the concept of personal assistance here.
As we've seen, the better the task is understood (through its description, expectations, and context), the better the result.
Well, ChatGPT benefits from getting to know you:
If you have regular requests, you don't need to ask the same things again or re-specify the context each time: the tool remembers and will respond to your request consistently each time.
If you have a niche subject, then all the context of your conversation can provide even more context for your questions.
For example, I've had a Gardening Assistant conversation for a year. <
If I ask it "when should I sow my peppers," it won't just give me a date; it will give me a contextualized response like "in February, before your tomatoes, even though they are from the same family, and the peppers will need more warmth than your tomatoes"; all because we've already talked about my tomatoes before.
What I've managed to do with ChatGPT professionally
Translate texts using context and a certain tone
Having recently migrated my website, I decided to make it bilingual (30% of WorkMeTender's missions were carried out with clients outside France, with a French website...).
That means more than 70 blog articles to update and translate. 😱
And there, ChatGPT is incredible because I can ask it to keep the same tone as the original article. I always specify the target audience of the article and prefer to specify my tone with adjectives to ensure my expectations are met, but again, if I use a "long conversation," I don't need to re-specify, this assistant knows my expectations for translating articles from my blog.
Note, the translation is not perfect and always requires proofreading, but it's more personalized than Google Translate and faster than translating my own article myself.
Outcome: Great results for this use.
Obtain synonyms
While it's very easy to obtain synonyms online, they are rarely contextualized. With ChatGPT, one can ask "give me a synonym for such word, knowing that I want to express such phrase..."
Outline ideas that I've formulated
When I need to clarify ideas that are on my mind, visually structure them, or even to challenge them to ensure they are understood, I present them to ChatGPT.
On a simple basis: if ChatGPT can reformulate my idea showing that it has perfectly understood it, it means the people I work with at my clients, or the people who attend my trainings will understand this idea in its context even more. Developers might say ChatGPT is my Rubber Duck.
On a few occasions, ChatGPT has even suggested tables or diagrams to explain an idea visually.
Outcome: Very pleasantly surprised.
Analyze documents or articles
A lot of content is published each month, and part of my job is also to stay informed about what's happening. But generally, the percentage of published articles that are really interesting (for my expectations) rarely exceeds 10%.
This view may seem demanding, yet there is a lot of noise, increasingly so since AI allows for effortless (and soulless, for that matter) content creation. It is therefore interesting to have a first filter, and I ask ChatGPT either to summarize an article for me or to extract 5 key points.
If what it tells me is interesting, I hurry to read the full article. If what it tells me seems to have little added value, I move on.
Outcome: I consume much less "short" content like articles thanks to this filter, which leaves me more time for "long" content like books.
Note that it is also possible to summarize YouTube videos these days with other tools, however, I haven't tested it yet: I use YouTube content more like podcasts and news on topics I follow particularly, and I appreciate this lively aspect.
Generate illustration images
Since it is directly connected to Dall-E, OpenAI's image generator, and Dall-E has been updated to version 3, we can generate images directly from ChatGPT by describing them.
It's extremely useful when you're a small structure that produces a lot of content, and you need very specific visualizations to illustrate blog articles (like here) or training materials.
The fact that image generation is linked to the conversation makes the process extremely easy, which can lead to simple modifications, for example, "make the image more joyful" or "move this object to the top of the image," etc...
One can obtain much better results with other image generation tools, in my opinion, such as Midjourney or a StableDiffusion trained on its own models, but Dall-E 3 through ChatGPT meets my needs for illustration.
What I HAVE NOT managed to do for working with ChatGPT
Generate relevant approach messages:
ChatGPT offers the possibility to create agents, which will be trained for a very specific purpose.
I then wanted to create an agent that, when given a candidate profile and a job offer, could generate a personalized approach message in my way. I mean a quality message, not a generic approach that could fit other profiles; I always go for hyper-personalization (and it works for me).
So, I anonymized hundreds of approach messages I had written, and fed them to this ChatGPT agent. I then trained it to understand my tone, the type of information I generally emphasize, my greetings, and the desired number of characters.
I finally set it to ask for a job offer and a copy of a profile before generating an appropriate and relevant approach message.
Outcome: for now, it's too much to ask.
I managed to get satisfying messages, but too irregularly (1 out of 4), and I couldn't get it to respect a character limit. Or as soon as I reminded it of this criterion, it would forget one or two others from the request.
After that, yes, you can always ask ChatGPT to write something better than "Are you open to the market?" but my expectations in terms of writing approach messages are very high, as in sourcing, I aim for very high response rates too.
So for message generation, in my opinion: still too early. However, we are not far from it, I am convinced!
Source on LinkedIn
ChatGPT is very disappointing when it comes to surfing websites in real-time, as it is considered a robot by said sites, and gets easily blocked.
If a conversational AI tool manages to interface with a social network's messaging one day, it will be both incredible and terribly dangerous.
Making jokes:
Yes, professionally, depending on the tone one uses, one may sometimes wish to add a touch of humor.
ChatGPT's jokes weren't really funny.
Besides, it shows well what it lacks: spirit.
However, I got results that could be qualified as amusing when asking it to rewrite texts "in the manner of" (less relevant for my professional uses).
So my interpretation is that it is capable of transforming, but not of creating. Again, the result depends on the richness of the information that is at its disposal. Example of prompt: Rewrite the fable of La Fontaine "The Ant and the Grasshopper" in the style of Steve Carell.
Bonus: what I managed to do with ChatGPT in my personal life
Have a Gardening Assistant
I'm lucky enough to be able to build a new vegetable garden in my yard, I always ask my questions on this topic in the same conversation.
It goes from "calculate the volume of soil I should put in this container" to "remind me of the minimum recommended distance between two tomato plants."
It's simple, and it works well, and ChatGPT and I can refer to an entire year of conversation if needed.
Have a DIY Assistant
It's a bit like my Gardening Assistant, I ask a lot of questions to ChatGPT for the work I do at home.
What's quite crazy is that ChatGPT is capable of challenging me when I present my ideas.
It is able to point out risks on which I did not explicitly solicit it (e.g., "it's a good idea to use shuttering boards to make vegetable containers, however, think to protect them with linseed oil to waterproof them.")
Obviously, I'm not going to entrust it to redo the plan of my electrical panel, but for many simple tasks, or ideas for fitting out, I find its answers quite comprehensive and, surprisingly, satisfying.
Find products online
Sometimes, it's not for lack of searching, I can't find what I need.
Nothing in store, too many eCommerce sites, when it's too specific, it's like looking for a needle in a haystack.
But ChatGPT has been able, if not to do my shopping for me, to give me product references corresponding to my need that are not easily found in the general public, I then just had to search and order the product references directly.
Example of prompt: I need to manufacture XXX and fix it on YYY, in order to do ZZZ. What materials do you recommend? Give me specific product references, and if possible, pricing information.
Create exercises for children
I'm lucky that my eldest son often asks me for exercise books during the holidays (especially calculations), but making them by hand always takes me time, because the goal is that it's adapted to his level while challenging him so he's happy to have learned new things. With ChatGPT, I can create them on the fly and tailor-made, specifying the expected level, the format, and the purpose of the exercise.
And many other things:
Develop Role-Playing Game campaigns, create mini-video games with children, develop children's creativity by bringing their stories to life, find cooking recipes based on a list of ingredients...
And you, what's your experience with ChatGPT?
Have you managed to do other things? Tell us!