ChatGPT, me and myself
Generative AI systems have arrived. They are shaking up the balance and enabling the creation of new tools. This is important. Let us explain why.
A parrot?
Chat stands for chatbot or conversational agent.
GPT stands for _Generative Pre-trained Transformer.
It's an artificial intelligence system capable of producing text.
There are two hilarious jokes in french about ChatGPT.
Chat means cat.
People love cats, as you may know.
So this is funny.
GPT looks like I farted.
Kids find it funny as well.
So ChatGPT sounds weird in french.
You have to deal with that.
Generative system
Talking about AI is a problematic shortcut.
As you know: AI does not exist.
These systems are not really intelligent in the human or animal sense of the term, because they don't think. They don't understand.
They produce text.
Or images, in the case of MidJourney.
These programmes are (perhaps) useful, but not really intelligent.
Stochastic parrot
Some researchers describe ChatGPT as a stochastic parrot.
This means that it repeats words randomly.
In practice, this is a bit wrong.
To put it quickly, ChatGPT assembles bits of sentences by matching them to a context that resembles what it has learnt.
The content is therefore generated without any internal reasoning.
This is the strong limit.
The image of the parrot is not a bad one, however.
You could also talk about exquisite corpse.
Who cares?
ChatGPT is still an improved toaster.
Automatic generation systems don't understand anything.
They're just ersatz intelligence.
Yes.
Of course they are.
But what's the point?
These systems are already shaking up the visual creation market. Every cloud has a silver lining: the intellectual property of digital objects is going to have to change.
Are we using them?
At least we're asking ourselves the question.
For visual content, it's settled: we use it.
For quick overviews of a subject, we use it.
For data aggregation and compilation, we're trying.
For B2C interactions, we're also trying.
So we're wondering.
...
As you might have guessed, the consultancy and strategy firms are sweating bullets. I think they're right.
OpenAI
They are the designers of ChatGPT.
They have links with Microsoft.
I don't know what to make of them.
In any case, they offer an API to their conversational agent.
For a fee.
It allows you to test these systems quickly.
It's great.
Hugging Face
This company was founded by three french guys.
But based in the USA.
They offer freely accessible language models (LLM).
Including Meta's Llama LLM model.
Soon those of Google.
As well as aggregated data sets.
This makes it possible to recreate systems equivalent to ChatGPT ...
... more or less.
How do you do it?
In practice, the ChatGPT-like market is still very new.
It's still in its infancy.
The technology is very new, and the community has little experience of it.
You have to be careful with glowing announcements.
But there's nothing to stop you taking bets.
Code & data
The model code is available. Or almost.
The training data is available. Or almost.
Generative systems can be parameterised. Or almost.
The maturity of all this remains rather low for a production run.
The work will take hundreds of days, at the very least.
But it's worth a try.
Lots of companies are going to do it.
Try it!
A lot of people want to give it a go.
It's understandable.
Devs capable of setting up these systems are extremely rare.
For the time being.
Shall we tell you how?