Nov. 2 (Portaltic/EP) –
Google has announced that it is developing a tool for creative writing capable of generating text and powered by a self-developed Artificial Intelligence (AI) technology known as Language Model for Dialogue Applications (LaMDA).
The company announced LaMDA at its 2021 annual developer event, when it featured it as an AI based on the Transformer neural network architecture created by the department of Google Research in 2017.
Now Google has taken stock of its latest advances in AI and has made several announcements regarding this technology, as a project with which it hopes to build a model that supports the 1,000 most widely spoken languages in the world.
“This will be a task of many years,” the technological giant has commented in an entry on your blogwhere he recognized that he is already taking “significant steps” and where he recalled that his “more advanced language models are multimodal, which means that they are capable of unlocking information in many other different formats”.
As part of this initiative, it has developed a universal voice model (USM), which trains in more than 400 languages and you have also referenced a recent announcement, when you indicated that Gboard already integrated voice dictation support for nine more African languages.
Another of the products that Google is currently working on is called wordcraft, a tool that builds on its state-of-the-art dialog system LaMDA to generate AI-powered text.
The company has acknowledged that has created this system to “explore the future of writing” and, to prove it, it has had the participation of 13 professional writers, who have been able to use this system for eight weeks to write with LaMDA.
According to Google, this language model is trained to recognize a typed word in a text and predict the next most appropriate word to continue typing. Likewise, It has a section of possible phrases that depend on the activity of the author.
For example, if a user types a sentence and clicks the ‘Retype this sentence’ button, the web application opens and displays a text field in which the user can describe how they would like the sentence to be rewritten.
Thus, you can choose the right tone for your text, so that, for example, it is funnier or livelier or convey nostalgia or sadness. To proceed with the rewriting, Wordcraft uses LaMDA and the context of the writing and generates a modified sentence.
In addition to rewriting tasks, Wordcraft has text continuation options and also integrates a control that registers arbitrary indications. In this way, the author can ask this tool to have the program describe a subject or character or an action of this to generate more text.
Also, Google has integrated into this feature a ‘chatbot’ function in Wordcraft that allows you to generate a conversation about the story that is being written. In this way, authors can use this application to write as well as write as an assistantto encourage creative workflow.
It should be noted that the company is aware that there is still a long way to go with this tool and that it presents the usual problems of machine learning models, such as training data derived from “a static representation of the world and they don’t have an innate ability to overcome biases” in content generation.