Artificial intelligence (AI) technology has emerged that may help patients who cannot speak because of accidents or diseases. It’s a technology that analyzes the patient’s brain activity and writes what he desires to say.

For instance, when a patient desires to ask for water, brain activity is analyzed to provide you with texts corresponding to “I’m thirsty” or “I would like water”.

Even when a thirsty patient thinks of drinking a water bottle or drinking water, relatively than words or sentences to talk, he may provide you with a message saying, “I would like to drink water.”

This capability is made possible by combining functional magnetic resonance imaging (fMRI) technology with large language models (LLMs). Researchers on the University of Texas within the US used fMRI to scan the brain activity of three subjects as they told a lengthy story.

A language model was then used to seek out patterns on this scanned image data. A big language model is about finding patterns in data and finding the almost definitely next word.

Nevertheless, the research team used a technique to discover the gist of the vast patterns in the information, relatively than pairing words individually. Because of this, we developed an algorithm that turns patterns into words in real time. It is called ‘Semantic Decoder’.

The activity of our brain is stuffed with a sort of coded signal, and it is claimed that it is feasible to decipher it with a language model.

Particularly, the research team revealed that after they applied the ‘semantic decoder’ while showing a brief animated video to the test subjects, a text explaining the video was created in real time.

It signifies that this algorithm has deciphered the activity of the brain that interprets the meaning after seeing the video. The researchers rated this tool pretty much as good for getting a gist or outline.

We are able to consider a word or sentence in our head, but we also can imagine an easy image or motion, and this can be a story that AI can read, the researchers explained.

Is it a bit scary at this point? In truth, the danger that this technology could be used to steal people’s thoughts has been identified. Nevertheless, there is no such thing as a have to panic because it is just a technology at the extent of confirming its potential.

Next, we’ll inform you in regards to the technology trends.

■ Nvidia released ‘DGX H100’, a supercomputing system for data centers. This method is a module that mounts eight ‘H100’ GPU chips and connects them to 1 through ‘Nvidia NVLink’.

DGX H100 can speed up exascale operations by connecting as much as 32, or a complete of 256 based on H100 GPUs. Dedicated transformer engine able to handling LLMs with trillions of parameters.

■ Microsoft (MS) and Google, that are in a search war, had a confrontation while announcing their first quarter results. Regarding the tendency to mix search and chat after the looks of ChatGPT, MS CEO Satya Nadella said, “A ‘generational shift in search’ is going down through chatbots.”

Google CEO Sundar Pichai, then again, emphasized that “it’s just one among many changes in search practices,” adding, “We all know that billions of individuals trust Google for its accuracy.”

■ Inflection AI, a startup founded by experts from MS and Google, has launched a chatbot ‘Pie’ that focuses on human interaction, unlike other AI chatbots corresponding to ChatGPT or ‘Bard’. Talk like an actual friend, asking how you’re feeling and giving comfort.

Inflection AI was founded by Reid Hoffman, co-founder of LinkedIn and director of OpenAI, and Mustafa Suleiman, co-founder of DeepMind, after their respective resignations in March. LinkedIn is a subsidiary of Microsoft, and Deep Mind is a subsidiary of Google, and their meeting was a hot topic even before the corporate was established.

■ The Ministry of Science and ICT will release 310 learning datasets sequentially through the ‘AI Hub’ by the tip of July. Corporations, researchers, and developers, in addition to anyone enthusiastic about AI development, can utilize learning data through the AI ​​Hub.

When the opening of roughly 1.5 billion pieces of information from 310 types is accomplished this time, AI hub users will find a way to utilize a complete of 691 kinds of data and roughly 2.6 billion pieces of information. Particularly, considering that this yr is a large-scale opening, data that has undergone quality verification will probably be opened first in order that users can quickly utilize plenty of data.

Next, we’ll inform you in regards to the major trends within the industry.

■ Newsguard, a fake news monitoring group, revealed that there are about 50 sites that randomly create fake news with generative AI corresponding to ChatGPT. With names like ‘News Live 79’ and ‘Each day Business Post’, they spread fake news, lifestyle suggestions and celebrity news.

These are sites which might be decorated like news media for the aim of promoting revenue, after which create and post content that may generate traffic through AI. is hitting

■ An AI expert named Santiago Valdarama created and presented a robot dog that may understand and reply to human speech. Boston Dynamics’ quadrupedal robot ‘Spot’ was equipped with ChatGPT and Google’s voice conversion AI to know and reply to human speech.

“The robot accumulates an enormous amount of information during missions, but there was no strategy to check it immediately,” said Valdarama. “You need to use ChatGPT to envision the mission results or contents.”

■ Databricks integrated ‘Apache Spark’ data frame to make it easier to make use of Hugging Face’s dataset. It makes it easier for users to generate high-quality data sets for AI model development.

Apache Spark DataFrame is an open source platform for facilitating the work of the large data processing a part of the large data pipeline that collects, stores, processes and manages data. It performs tasks corresponding to data evaluation and preprocessing, data processing and extraction, and ultra-high-speed parallel processing.

■ Medical AI Solution Coreline Soft, a domestic company, is entering the US market. The corporate signed a contract with Fullmons, a world leading company in the sphere of chronic obstructive pulmonary disease (COPD), to produce ‘Aview COPD’, an automatic lung image evaluation software.

Fullmons is an organization based in Redwood City, California, USA, and Neuchatel, Switzerland, and is legendary for its respiratory disorder treatment equipment, including the ‘Zephyr Valve’, a tool for treating dyspnea related to severe emphysema. ‘Aview COPD’ supplied by Coreline Soft to this company is an answer that completely automates the image evaluation algorithm required for COPD diagnosis.

Jeong Byeong-il, member jbi@aitimes.com