For AI, ChatGPT is a real leap into the future –

For 5 days I have been appalled and amazed at the same time.

The launch of chatGPT was unclear. I couldn’t figure out if it was a small update to the GPT-3 artificial intelligence algorithm, launched in 2020 by OpenAI. Or on the contrary a huge secret update with the version GPT-4 announced for next year. It’s neither apparently. Just a little “fine-tuning”, version 3.5. I had come across it on my LinkedIn, like you probably recently without really understanding, wondering if I wasn’t just 2 years behind on the appreciation of GPT-3, the most successful AI to date .

How it works? Basically, you ask a question and get a well-structured, essay-worthy answer with great relativity and caution in posture. You can also ask for a text to be written (example), called prompt. Or rephrase this article in its entirety, for example. I tried, but I still preferred to redo it myself…

Five days of testing

Test if it hasn’t already been done and if the servers are available: one million users in five days. 50 million by Christmas? I made dozens of screenshots on custom phrases for my friends. For my daughter and my partner, I showed them examples of texts in medicine, my son the harmful effects of too many video games, my Instagram network some tips on… Instagram, my clients the challenges of hybrid working, my former associate the difference between a feedback and a comment, etc.

I’ve seen people ask AI to make startup-specific business plans who sell software online (SaaS)define 3 years of completely credible objectives, ideas to present to Elon Musk to avoid being fired or write great ads for LinkedIn based on just a few words. The more I tested, the more I trembled.

This AI creates on the fly, without having fun and without understanding the impact on our emotions. She still takes gloves, tries to avoid biases and trick or immoral questions like “how to rob people in a car”.

Pitfalls and mistakes

But it gets tricked too, with some people using tricks by adding video game context to get a limitless response. And she is already banned, for example by StackOverflow because the computer codes it generates also contain errors. And developers integrate it with systems which imitate the human browsing a web page. “What could go wrong?”

Then I saw other amazed people arriving on my LinkedIn, a bit like after a bomb on tv. I felt less alone. The Guardian was the first to note the surprise of the academic community: “AI bot ChatGPT stuns academics with essay-writing skills and usability”.

“Clickbait” a friend replies. Whatever, it was according to the history of Google News the first major public media to talk about it, 3 days late anyway. Then it’s the avalanche on Techmeme, LinkedIn & Co. Human consensus, slower than the algorithm but so rich, and irreplaceable, don’t worry.

Some hot reactions: “Google is dead”, “ I just had a 20 minute conversation about modern physics…”, “What dishonest”, “Still not settled the problem of bias”, “we are not far from a Dangerous AI”.

Always the topic of trust!

In many ways, the flaws in ChatGPT’s seemingly brilliant responses trick the human into understanding that an eloquent person is not automatically trustworthy.

Let’s take the glass half full hypothesis: it is difficult for us to measure which sectors will be directly affected by such a tool. It is beyond our imagination. It is also complicated for us to evaluate the possible cocktails.

Reserved for subscribers

This article is for subscribers only.

Already subscribed?
To log in

We wish to thank the writer of this post for this remarkable content

For AI, ChatGPT is a real leap into the future –

Discover our social media accounts as well as other related pages