2022 review of main AI news
Here comes the end of the year, according to tradition, at the end of the year, the results are summed up and plans are made for the next one, as a summary, I would like to consider the most interesting news from the world of artificial intelligence that we have come across this year.
OpenAI released a smart chatbot ChatGPT
On November 30, OpenAI released ChatGPT, a prototype AI chatbot. It is based on the language model of the GPT-3.5 family. The chatbot is trained on text and code, and has undergone reinforcement training with human feedback. After the company put the chatbot in preliminary demo access, conversations with ChatGPT went viral on the network. He became famous for his detailed and well-formed answers, although sometimes false.
ChatGPT will easily create a letter or review for you, help you copy the article and even write a post on social networks.
It should be remembered that ChatGPT perfectly forms a complete text that looks professional and high-quality, but often the answers it provides are not correct. Be careful and check the information offered to you.
ChatGPT is a great product that can make many routine tasks easier for you, definitely one of the best releases of the year.
Stable Diffusion, DALL-E 2, Midjourney
The year 2022 has become very rich in the releases of generative neural networks, I think many of you have already tried to generate at least a few images using one of the above networks, and it really works well. These networks caused great resonance among artists and representatives of creative professions because they made it possible for more people with the idea to generate beautiful and realistic images in different styles. Pessimists, to whom the majority of artists and photographers belong, believe that the appearance and distribution of these tools will lead to the disappearance of representatives of creative professions. Optimists believe that humanity has received new tools that will allow humanity’s creativity to reach a new level.
In my opinion, the optimists are right and we will see rapid development of image generation technologies in the future.
War with Unstable Diffusion
The release of Stable Diffusion provoked one interesting phenomenon — the emergence of an “improved network” Unstable Diffusion is completely similar in structure but without restrictions on the generation of any content. This was immediately taken advantage of by some people for whom it was important not to have restrictions on content, of course, the network was also partly used to generate erotic and other unacceptable content. In turn, this has led to the activation of the AI ethics community, a story that continues to develop, and therefore we will witness further developments in the field of AI regulation and its application in the future.
Developments of Google-2022: synthesis of video by text and editor of artistic text
Google even has not one, but two models that can generate videos from tips. Phenaki can create videos up to 2 minutes long upon your request. Its main drawback is poor image quality. Instead, the Imagen Video model produces short, high-resolution clips. Google combined both models into one to take advantage of their advantages. The combined model can create long, high-quality sequential videos.
The corporation also has the development of Wordcraft, which is based on LaMDA (Language Model for Dialogue Applications). Her task is to imitate speech patterns by studying vocabulary on the Internet.
Wordcraft is a tool for creating fiction with artificial intelligence. It is intended for creative writing, storytelling, and experimental fiction. Unfortunately, Wordcraft, like previous Google tools, is not publicly available. Google does not provide access to these services.
Users can rewrite sentences and have the AI change them, such as “make it funnier.” They can also ask him to continue ideas or improve existing ones. In addition, Google has added a feature they call freeform tooltip. Given the context, the user can ask Wordcraft to generate a prompt (for example, “tell me what the old man felt about the flowers” if your story already mentions this character).
To test Wordcraft in the real world, Google created the Wordcraft Writers Workshop. They invited 13 professional creative authors: screenwriters, poets, educators, etc. to evaluate the tool. Together they created a story. As far as I know, this is the first work in this format.
Fail of the Year: Galactica Scientific Algorithm
In November, Meta announced a new large language model (LLM) for science and called it Galactica, a nod to Isaac Asimov’s Encyclopedia Galactica. The website says it’s “a model that can store, combine, and reason about scientific knowledge.”
The algorithm’s purpose is no doubt noble: the company conceived Galactica to “organize science” as humanity knows and stores more and more knowledge about the world. However, the total percentage we can absorb is rapidly decreasing.
Galactica is a family of language models trained on 60+ million high-quality scientific documents (articles, textbooks, encyclopedias). Meta claimed that Galactica can “summarize academic literature, solve math problems, generate Wiki articles, write scientific code, annotate molecules and proteins, and more.”
And this seems to be great, but the model has been criticized by scientists. For example, Simon J. Greenhill, a professor at Auckland University’s School of Biological Sciences, said: “A great service to paper mills, fraudulent plagiarists & cheating students everywhere.”.
Conclusions
Throughout the year, there were numerous advancements made, particularly in the field of image generation. It is likely that many important news were not included in this review, it will be great if you share any news that had a significant impact on you it his year in the comments section. As artificial intelligence continues to evolve and enhance people’s lives, we are committed to supporting its development and advancement.
And if you need help with your computer vision project, AIKOLO will help you solve your problems in the shortest possible time.
