Google on Tuesday invited people in the United States and Britain to test its AI chatbot, known as Bard, as it continues on its gradual path to catch up with Microsoft-backed ChatGPT.

Bard, ChatGPT and other similar artificial intelligence apps churn out essays, poems or computing code on command and have taken the world by storm as the biggest new thing in tech since the advent of the iPhone.

Google CEO Sundar Pichai told staff that after testing Bard with 80,000 Google employees, the chatbot would be tested with the public in the United States and Britain as a “first step” before going out to more countries in other languages.

“As more people start to use Bard and test its capabilities, they’ll surprise us,” Pichai said in a memo to staff seen by AFP.

“Things will go wrong. But the user feedback is critical to improving the product and the underlying technology,” added Pichai, who had faced some criticism within the company for rushing to catch up with Microsoft.

In the launch, people wishing to play with Bard can sign up on a waiting list at bard.google.com website, distinctly separate from the tech giant’s search engine.

“We’ve learned a lot so far by testing Bard, and the next critical step in improving it is to get feedback from more people,” Google vice presidents Sissie Hsiao and Eli Collins said in a blog post.

As exciting as chatbots can be, they have their faults, Hsiao and Collins cautioned.

‘Constantly learning’

Google has so far proceeded more carefully in its rollout of generative AI to consumers, in contrast to Microsoft’s choice to swiftly make the products available despite reports of problems.

ChatGPT’s OpenAI is backed by Microsoft, which earlier this year said it would finance the research company to the tune of billions of dollars.

Asked by AFP how its product was different from ChatGPT, Bard said that unlike its Microsoft-backed rival it was “able to access and process information from the real world through Google Search and keep my response consistent with search results.”

The bot also underlined that it was still “under development, while ChatGPT has been released to the public. This means that I am constantly learning and improving, while ChatGPT is likely to remain relatively unchanged.”

OpenAI recently released a long-awaited update of its AI technology that it said would be safer and more accurate than its predecessor.

Much of the new model’s firepower, known as GPT-4, is now available to the general public via ChatGPT Plus, OpenAI’s paid subscription plan and on an AI-powered version of Microsoft’s Bing search engine.

Microsoft has said that its quick adoption of generative AI has seen usage of its Bing search engine increase in recent weeks, but it is still a clear underdog to Google, which captures about 85 percent of the global search engine market.

© 2023 AFP

MechSE Associate Professor Kyle Smith and doctoral student Md Abdul Hamid recently published an article in the Journal of Power Sources.

Their study, “A bottom-up, multi-scale theory for transient mass transport of redox-active species through porous electrodes beyond the pseudo-steady limit,” demonstrates that conventional theory approaches underpredict the power capacity of flow batteries such as those used in electricity production from renewable energy sources.

“We have discovered a new means by which to understand how convection occurs inside of reactive porous media,” Smith said of their theory, which proposes introducing certain frequency-dependent transfer functions to up-scale mass transport occurring in the microscopic pores of electrodes. Although transfer functions are routinely used as mathematical tools in control theory, they have never before been applied in this context or derived in this manner.

The two began formulating their theory before the COVID-19 outbreak, making their publication a long-anticipated success. Their theory sheds new light on familiar mass and heat transfer principles—they introduced a spectral Sherwood number, which is a type of transfer function, to extend the film law of mass transfer to transient conditions. Similarly, a spectral Nusselt number extends Newton’s law of cooling for convection heat transfer. The two formulated the embedding of transfer functions into an up-scaled model to obtain the time-domain response of flow batteries.

“We have uncovered a new understanding of certain non-dimensional parameters that are ubiquitous in convection heat/mass transfer,” Smith said. “For the first time to our knowledge, we have extended these ideas from their conventional application in time-invariant, or steady state, settings to transient settings in a manner that accounts for changes in the microscopic dynamics that result from transient cycling.”

This work is also meaningful to chemical, civil, and petroleum engineering communities, which have explored approaches to understanding mass transfer in other porous materials. “Their approaches were not previously applied to the electrochemical systems that are the subject of our work,” Smith said. “However, we have developed a relatively straightforward approach to model these effects by using a formulation that starts from the detailed microstructure and up-scales its effects for use in macroscopic scale models.”

Indeed, the team’s theory demonstrates that flow batteries can be operated at higher than their limiting current for short periods of time, which suggests that cheaper and lighter batteries can be designed for these cycling conditions.

“These findings not only provide a better modeling approach for accurate prediction of an existing flow battery’s rate capability, but also provide guidelines toward more efficient designs, operating schemes, and materials,” Hamid said.

Smith and Hamid intend to apply their theory to different microstructures for porous electrodes for various energy and environmental devices using electrochemistry. Their next steps also include extending the theory beyond the range of conditions demonstrated in their publication.

Recognizing when senior citizens are at risk in the home or helping them find misplaced objects they presumed lost: The technology developed in the successful OMNICONNECT project can help people lead independent lives for longer. The researchers of Fraunhofer IZM have integrated a miniature radar system into an LED ceiling light that can track and recognize movement patterns and locate people or objects in a room.

Four radar modules are combined to ensure full 360° coverage. AI algorithms work through the collected data to identify whether a person in that range has suffered a fall, with the radar system being able to track more than 30 people in spaces of up to 1600 square feet with an angular resolution of 12°. And for added object detection capabilities, passive transponders were developed to work with the radar system.

Supported by the Ministry of Education and Research, the OMNICONNECT project has brought together researchers from the Fraunhofer Institute for Reliability and Microintegration IZM with partners from Berlin and Oldenburg.

Their mission: To develop a miniature radar system that can detect medical emergencies or other situations that require a caregiver’s intervention, while respecting the privacy of the users. By contrast to cameras or CCTV systems that record images, the system only tracks patterns of movement. It is integrated into a ceiling light, making it easy to install, unobtrusive, and easy to accept for the people living with it.

The system works with AI-supported radar modules and passive transponders that can be placed on different objects, thus for the first time enabling the tracking of movement and location with a single radar system. The transponders work as frequency-specific radar targets, with each resonating at a known frequency. This turns them into uniquely identifiable beacons that the system can scan for and whose precise location can be calculated by checking the time needed for the signal to bounce back.

The data is processed on site by a field-programmable gate array or FPGA, a tiny computing system with integrated processor. It shares the object and motion data with an AI-based system that can recognize movements and patterns, developed by the computer science researchers at the OFFIS Institute of the University of Oldenburg. To control the app and identify specific options, a special app interface was built by HFC Human-Factors-Consult GmbH.

The demonstration unit designed and constructed at Fraunhofer IZM has undergone live trials and proven its ability to detect an object’s position with an accuracy of five centimeters in a ten-meter range. The demonstrator is currently being used by the project partners in several possible use cases. The data it collects can be used to produce certain movement patterns of typical behaviors and, with enough meaningful data available, feed into possible assisted living applications or be used to recognize certain incidents. The position data could be used to check whether the detected person is feeling well or whether a medical or care intervention is indicated.