Al summarizing documents in modern business English and ancient Greek
March has brought an interesting insight into real-life applications of the language processing abilities of modern machine learning solutions.
This edition also offers a glimpse into users’ willingness to support their jobs with AI-powered tools.
You.com – AI-powered search engine launches GPT-3 based writing assistant
You.com is a new player in the Google-dominated market of search engines. What makes it stand out from the crowd is that it uses artificial intelligence to parse through multiple websites and display the desired result directly in the search results. When combined with apps that integrate with the engine it aims to enable the user to complete the information search process without leaving that particular window.
To enhance the process of preparing reports from the searched documents and summarizations of gathered knowledge, the search engine has launched an automated writing assistant. The user enters the audience, the topic, and desired length of the text, and YouWrite does the rest – as well as an OpenAI GPT-3 model can.
The service will initially be offered for free, but the company aims to bill its most frequent users in the future.
More details about the tool are available on the company’s blog.
Deepmind deciphers ancient texts using AI
After solving the mystery hidden in the way proteins are folding and tackling the problem of overheating computers in Google’s server farms, Deepmind has found a new area of interest to disrupt – ancient Greece history.
The new model called Ithaca aims to restore ancient Greek texts that are missing fragments and which pose challenges to properly dating. The system uses the Natural Language Processing approach to identify the most commonly used words and, based on previously analyzed texts, deliver the missing parts.
According to Deepmind’s blog post, the model has reached 62% accuracy in restoring damaged texts, and 71% accuracy in identifying the original location of the document.
More details about this research can be found in Deepmind’s blog post and a research paper published in Nature.
Gartner: 70% of consumers are willing to use AI to support their job
Gartner has recently delivered a report about the ways US consumers are willing to support their work with the help of AI. According to the report, there is not much fear, but rather curiosity and hope regarding the way machines will impact the way people work.
The majority (58%) would like AI to reduce the number of mistakes they make, with support in problem-solving (57%), and with simplifying processes, and discovering information (both 56%) also being popular answers.
Yet human workers are still willing to stay in control of the ways their data is being used to train AI-based solutions.
Interestingly, users are least interested in applying AI in the performance of physical tasks (25% pointing out that AI needs to do nothing on his or her behalf) or in supporting the keeping of the rules related to one’s own work (26%).
The full report can be found on Gartner’s webpage.
Google introduces auto-generated summaries in Google Docs
The volume of text increases every day, especially if one aims to stay on top in his or her area of expertise. On the other hand, though, not every text is worth reading. Writing summaries can be a great way to help readers manage their time and attention.
On the other hand, writing a good summary is an art in itself and can be extremely time-consuming, especially for a writer who hadn’t encountered the text before.
To facilitate this process, Google has delivered a summarization engine embedded in Google Docs. With it, the user can ask an AI to provide a short text that has all the important information squeezed in and delivered straight to the reader.
The system is currently available only through paid plans of the Google Workspace. More details can be found in the Google blog post about this new feature.
OpenAI introduces Dall-e 2
The organization behind GPT natural language processing models has introduced the next generation of Dall-e neural networks. This particular network aims to understand the relationship between words and images. The network is capable of producing an image from a written description using multiple styles, for example, a cave painting, kid-style crayons, or a Johannes Vermeer-like painting.
The neural network can also apply realistic edits to existing images, for example inserting a flamingo into an image. Depending on the context depicted, it can become a toy indoors or a real bird in the sky.
The details about the new network can be found in this paper published by OpenAI and on their website.
