Retrospective 2022 – The year of generative AI

The year 2022 will have been largely marked by one technology: generative AI. These “creative” intelligences like Dall-E and ChatGPT have made a quantum leap in artificial intelligence and will transform our daily life in business in 2023…

Many CIOs and experimenters of new technologies have, in 2022, toyed with generative AIs. They are called Dall-E, Point-E, Stable Diffusion, MidJourney, CoPilot, or ChatGPT. Behind these barbaric names hide new Deep Learning technologies (GPT, GANs, etc.) capable of generating text, programming code, images, videos and 3D objects from scratch.

They were sometimes a little oversold. Often we tried to make them do what they were not designed for. Like ChatGPT, which many curious people see as a truth teller, the google of the future, a technological bible, a cybercriminal weapon when it is only an AI storyteller, educated to handle languages and words in order to maintain conversations even if it means inventing and not to tell verified and controlled truths.

But in 2023, these new AIs – still very largely experimental – will infiltrate the daily lives of developers, journalists, editors and creative people to allow them to find inspiration or simply to be more productive.

A quick tour of the AIs of the year…

ChatGPT

Give honour where honour is due. It is the AI ​​that will have caused the most ink to flow. We gave it extensive coverage in our paper: The amazing “ChatGPT” AI: what CIOs absolutely need to know in 10 points
This AI demonstrates the spectacular progress made to transfigure human-machine interactions and enable a natural dialogue between human beings and AIs.
As we explain in our article, ChatGPT has not been trained with certified databases, but to manipulate languages. His goal is not to answer correctly, but to tell stories. This AI has a very tangential approach to truth. It can approach it and move away from it in the same sentence. But it can be of real service when it comes to reformulating sentences or ideas, finding titles, explaining technologies, illustrating technical concepts, producing stories.
An AI that we will talk a lot about in 2023, that’s obvious.
To discover it: chatGPT from OpenAI

READ ALSO :

Dall-E 2

Created by OpenAI, this AI surprised more than one specialist by generating images from words. Ask her “draw me a unicorn ridden by a cosmonaut and galloping on the Moon” and she will create an original work for you. Admittedly, sometimes this work can be very inspired by another. But it remains a ‘creation’ and not a direct copy-paste.
The service is already widely used since it generates 2 million images per day.
Microsoft has integrated it into its Microsoft Designer and is preparing to integrate it with BING and EDGE.
To discover it: OpenAI SLAB 2

Stable Spread

Often more artistic, more complex to access, but ultra-promising, Stable Diffusion is another service that creates images from a textual description. The service of Stability AI has over 10 million daily users. The service made a name for itself by winning an artistic prize!
To discover it: Stable Diffusion 2.1 Online demo or the official website DreamStudio

MidJourney

Competitor of Dall-E and Stable Diffusion, MidJourney is a tool increasingly used by artists to generate new universes. MidJourney has two million members and is used through a Discord server.
To discover it: Quick Start Guide – Midjourney

Point

This is the latest production from OpenAI, which will definitely have had a very active year in 2022, opening up these technologies to as many people as possible. Dot-E is a bit like Dall-E for the 3D world. It generates 3D objects or scenes (in Blender format) from a textual description.
To discover it: Point cloud diffusion for 3D model synthesis

READ ALSO :

Make-A-Video and Imagen Video

Inspired in particular by the work of Chinese universities, like CogVideo, Meta and Google have each unveiled the results of their research with services allowing the generation of video sequences from a textual description. Access to these services is currently restricted.
To discover : Make-a-Video by Meta and Image from Google Research


CoPilot and AlphaCode: the future of development?

AlphaCode of DeepMind (Google’s sister startup) and CoPilot of GitHub (which derives from the IA Codex of OpenAI) also caused a lot of ink to flow in 2022.
These tools make it possible to automatically generate lines of code from a textual description of the program.
These AIs are presented as the second member of the ‘pair-pogramming’ assisting the developer to make him more productive or detect possible anomalies in the code he has written.
However, these AIs – like the other AIs presented here – currently raise many debates around copyrights because they do not deliver the “sources” that inspired them to generate their lines of code. It would therefore violate the terms of the GPL, Apache and MIT open-source licenses which require that the names of the authors be recalled.
Last November, we learned that developer and lawyer Matthew Butterick was suing Microsoft as well as its subsidiary GitHub and its partner OpenAI for violation of open-source licenses by the AI ​​Copilot!
We will have to follow the evolution of this legal case in 2023. The consequences could affect all generative AIs…

To discover :
AlphaCode from DeepMind
GitHub CoPilot


READ ALSO :

We would love to say thanks to the author of this post for this amazing web content

Retrospective 2022 – The year of generative AI


Check out our social media accounts and also other related pageshttps://yaroos.com/related-pages/