[Chronique d’Alain McKenna] GPT-4, exponential jump after ChatGPT

The speed at whichartificial intelligence (IA) evolves is already crazy. Expected arrival in late 2023 of GPT-4, the next-generation system that brings interfaces to life ChatGPT AND tabletwill further accelerate the trend.

People who are already breathless from the advances in natural language understanding and automated generation in recent months are encouraged to take a breath, look in a dictionary and note down the definition of “exponential”. Because that’s pretty much what we’re experiencing these days: an improvement that seems to be faster and faster in computerized systems available to almost everyone.

Where will these advances stop? It seems difficult to predict. On the other hand, we should already think about the perverse effects that these systems will have on the daily lives of many people if they are not supervised very soon.

general artificial intelligence

Because the balloon must be immediately prevented from inflating at a certain burst level. Remember that, despite the latest progress, the AI ​​interfaces we’ve seen emerge in recent months aren’t perfect. They come a little closer to that other beloved computer research concept, that of an artificial general intelligence, or “AGI” as its acronym is in English, but they’re not yet connected.

Such a system, according to the accepted definition, would take the form of software that could be asked about almost any kind of problem and that would know how to find the right solution. Such an AI can in principle perform all the intellectual tasks that humans are capable of performing.

Of course, to achieve this, such an AI would need to be able to distinguish between what is true, and not just possible, and what is false, even if possible. Ideally, his creations, his works, should be authentic, and not just a collage of styles or inspiration obviously taken from others. This is most evident on the Dall-E image generator side, which is better at imitating existing artists’ pencil or brush strokes than developing its own visual signature.

That said, the track is already unclear: on certain Internet forums where visual artists, we started banning visitors whose creative style looks too much like an AI. The same may happen in schools later this winter, when the first big papers are marked and teachers may worry about reading papers that suddenly seem more researched than usual.

In other words, even imperfect, the AI ​​currently accessible via the Internet is already making its presence felt everywhere. As usual, hardly anyone was ready to see a seemingly ordinary website that radically changes their habits. Especially not governments.

It would be nice if a framework to limit the abuses and failures of tomorrow’s AI could see the light of day… today. And not in five years. Europe could adopt its own framework next summer.

It won’t be too soon, as the next Dall-E and ChatGPT upgrade of this world is expected later this year.

GPT-4

Going live in early December 2022, ChatGPT only took a week to shake up the entire internet. This web-based software is based on a machine learning model called GPT-3. It was, at the time, the latest model version of the “ Pre-trained generating transformer continuously produced and developed by researchers at OpenAI, the company behind this system.

In understanding and generating text or language, the GPT-3 system processes the information it receives from 175 billion different criteria. it is big! This is also a serious problem with this technology: it is extremely energy intensive.

However, this is nothing compared to what GPT-4, its next iteration, will be. According to what OpenAI executives said recently, GPT-4 can process information according to a number of criteria rumored to be between 1 and 100 trillion (or between 1000 and 100 trillion criteria).

This system will be much more powerful; he will almost read our thoughts!

He could better understand our intentions behind the questions we ask him, without them being obvious. He could also respond in a way that was even more natural, or at least better able to mimic human conversation.

OpenAI recognizes the limitations of its technology and can include safeguards in GPT-4 that will limit the AI’s ability to invent lies. This can take the form of reinforcement learning algorithms taking more into account user feedback.

Because, yes, interacting with humans allows the machine to improve. Taken upside down, this means that every time we have fun with Dall-E or ChatGPT, we contribute to accelerating the development of a machine that will undoubtedly one day replace us in many everyday tasks.

And this should happen in 2023…

To see it on video

Leave a Comment