Connect with us

Tools & Platforms

AI Flow by TeleAI Recognized as a Breakthrough Framework for AI Deployment and Distribution by Omdia

Published

on


SHANGHAI, July 11, 2025 /PRNewswire/ — AI Flow, the innovative framework developed by TeleAI, the Institute of Artificial Intelligence of China Telecom, has been recognized as a key role in the intelligent transformation of telecom infrastructure and services in the latest report by Omdia, a premier technology research and advisory firm. The report highlights AI Flow’s exceptional capabilities in addressing the edge GenAI implementation challenges, showcasing its device-edge-cloud computing architecture that optimizes both performance and efficiency as well as its groundbreaking combination of information and communication technologies.

According to the report, AI Flow facilitates seamless intelligence flow, allowing device-level agents to overcome the limitations of a single device and achieve enhanced functionality. The same communication network can connect advanced LLMs, VLMs, and diffusion models across heterogeneous nodes. By facilitating real-time, synergistic integration and dynamic interaction among these models, the approach achieves emergent intelligence that exceeds the capabilities of any individual model.

Lian Jye Su, Chief Analyst at Omdia, remarked that AI Flow has demonstrated sophisticated approaches to facilitate efficient collaboration across device-edge-cloud tiers and to achieve emergent intelligence through connective and interactive model operations.

The unveiling of AI Flow has also drawn great attention from the AI community on global social media. AI industry observer EyeingAI said on X “It’s a grounded, realistic take on where AI could be headed. ” AI tech influencer Parul Gautam said on X that AI Flow is pushing AI boundaries and ready to shape the future of intelligent connectivity.

Fulfill the Vision of Ubiquitous Intelligence in Future Communication Networks

AI Flow, under the leadership of Professer Xuelong Li, the CTO and Chief Scientist of China Telecom and Director of TeleAI, is introduced to address the significant challenges of the deployment of emerging AI applications posed by hardware resource limitations and communication network constraints, enhancing the scalability, responsiveness, and sustainability of real world AI systems. It is a multidisciplinary framework designed to enable seamless transmission and emergence of intelligence across hierarchical network architectures by leveraging inter-agent connections and human-agent interactions. At its core, AI Flow emphasizes three key points:

Device-Edge-Cloud Collaboration: AI Flow leverages a unified device-edge-cloud architecture, integrating end devices, edge servers, and cloud clusters, to dynamically optimize scalability and enable low-latency inference of AI models. By developing efficient collaboration paradigms tailored for the hierarchical network architecture, the system minimizes communication bottlenecks and streamlines inference execution.

Familial Models: Familial models refer to a set of multi-scale architectures designed to address diverse tasks and resource constraints within the AI Flow framework. These models facilitate seamless knowledge transfer and collaborative intelligence across the system through their interconnected capabilities. Notably, the familial models are feature-aligned, which allows efficient information sharing without the need for additional middleware. Furthermore, through well-structured collaborative design, deploying familial models over the hierarchical network can achieve enhanced inference efficiency under constrained communication bandwidth and computational resources.

Connectivity- and Interaction-based Intelligence Emergence: AI Flow introduces a paradigm shift to facilitate collaborations among advanced AI models, e.g., LLMs, vision-language models (VLMs), and diffusion models, thereby stimulating emergent intelligence surpassing the capability of any single model. In this framework, the synergistic integration of efficient collaboration and dynamic interaction among models becomes a key boost to the capabilities of AI models.

See AI Flow’s tech articles here:

https://www.arxiv.org/abs/2506.12479

https://ieeexplore.ieee.org/document/10884554 

AI Flow’s First Move: AI-Flow-Ruyi Familial Model

Notably, TeleAI has just open-sourced the first version of AI Flow’s familial model: AI-Flow-Ruyi-7B-Preview last week on GitHhub.

The model is designed for the next-generation device-edge-cloud model service architecture. Its core innovation lies in the shared intermediate features across models of varying scales, enabling the system to generate response with a subset of parameters based on problem complexity through an early-exit mechanism. Each branch can operate independently while leveraging their shared stem network for computation reduction and seamless switching. Combined with distributed device-edge-cloud deployment, it achieves collaborative inference among large and small models within the family, enhancing the efficiency of distributed model inference.

Open-source address

https://github.com/TeleAI-AI-Flow/AI-Flow-Ruyi 

About TeleAI

TeleAI, the Institute of Artificial Intelligence of China Telecom, is a pioneering team of AI scientists and enthusiasts, working to create breakthrough AI technologies that could build up  the next generation of ubiquitous intelligence and improve people’s wellbeing. Under the leadership of Professor Xuelong Li, the CTO and Chief Scientist of China Telecom, TeleAI aims to continuously expand the limits of human cognition and activities, by expediting research on AI governance, AI Flow, Intelligent Optoelectronics (with an emphasis on embodied AI), and AI Agents.

For more information:

https://www.teleai.com.cn/product/AboutTeleAI

Photo – https://mma.prnewswire.com/media/2729356/AI_Flow.jpg



Source link

Tools & Platforms

U.S. State Courts Cautiously Approach AI Despite Efficiency Promises and Staffing Crises

Published

on


A new survey of state courts reveals a striking paradox in the American judicial system: Even though courts face severe staffing shortages and operational strain, they remain reluctant to adopt generative artificial intelligence technologies that could provide significant relief.

The Thomson Reuters Institute’s third annual survey of state courts, conducted in partnership with the National Center for State Courts AI Policy Consortium, found that 68% of courts reported staff shortages and 48% of court professionals say they do not have enough time to get their work done.

Despite these pressures, however, just 17% say their court is using gen AI today.

Courts Under Strain

The survey, which gathered responses from 443 state, county, and municipal court judges and professionals between March and April 2025, paints a picture of courts under significant strain.

Seventy-one percent of state courts and 56% of county/municipal courts experienced staff shortages in the past year, with 61% anticipating continued shortages in the next 12 months.

This staffing crisis translates into demanding work schedules, with 53% of respondents saying they work between 40 and 45 hours a week on average, and an additional 38% working over 46 hours a week.

Perhaps most telling, only half of court professionals said they had enough time to get their work done.

These workload pressures are only getting worse. Nearly half of respondents (45%) reported an increase in their caseloads compared to last year and 39% said the issues they are dealing with have become more complex.

Meanwhile, 24% of respondents reported increases in court delays, compared to 18% who reported decreases.

AI Adoption Remains Limited

Against this backdrop of operational strain, the survey reveals a cautious approach to AI adoption that seems at odds with the technology’s potential benefits.

Currently, only 17% of respondents said their court was using gen AI, and an additional 17% said their court was planning to adopt gen AI technology over the next year.

This slow adoption occurs despite widespread recognition of AI’s transformative potential, with 55% of respondents rating AI and gen AI as having a transformational or high impact on courts over the next five years.

The survey found that AI and gen AI is the highest-ranking impactful trend, rated as transformational or high impact by 55% of respondents.

Court professionals clearly see the efficiency benefits AI could provide. Court professionals predict that in the next year, gen AI will help them save an average of nearly three hours a week, rising to nearly nine hours a week within five years.

The projected time savings could be substantial: Respondents estimate they will save an average of nearly three hours every week in the next year, growing to nearly six hours each week within three years and 8.8 hours each week within five years.

Barriers to AI Implementation

So what is keeping courts back? The survey identifies several factors contributing to courts’ cautious AI adoption.

Seventy percent of respondents said their courts are currently not allowing employees to use AI-based tools for court business, and 75% of respondents said their court has not yet provided any AI training.

There are also varied but significant concerns about AI implementation.

More than a third (35%) are worried that AI will lead to an overreliance on technology rather than skill, while a quarter have concerns about malicious use of AI, such as counterfeit orders and evidence. Interestingly, only 9% were worried about widespread job loss resulting from AI.

Budget constraints may also play a role in limiting technology adoption. The survey found that 22% say their budget for the next year increased, while 30% said budgets decreased, and 30% say budgets stayed the same.

Current Technology Landscape

While AI adoption lags, courts have made progress implementing other technologies. Most courts have adopted key technologies, including case management (86%), e-filing (85%), calendar management (83%), and document management (82%).

Video conferencing has reached near-universal adoption at 88%.

However, some technology gaps remain. Beyond gen AI, the most common technologies set to be adopted next are legal self-help portals, online dispute resolution and document automation.

Virtual Hearings Widely Adopted

The survey shows significant adoption of virtual hearings, with 80% of respondents saying their court conducts or participates in virtual hearings.

In more than 40% of all jurisdictions, virtual hearings are available for first/initial appearances, preliminary/status hearings and/or motion hearings.

Virtual hearings appear to improve court efficiency in some areas. 58% of respondents reported that virtual courts decrease failure to appear rates, and 84% reported that virtual courts increase access to justice.

However, the digital divide presents ongoing challenges. Nearly one in five respondents (19%) feel that the majority of litigants are experiencing decreased access to justice because they lack strong technology skills.

Court access for people with lower digital literacy and fewer technical support resources were ranked as the top challenges for litigants involved in virtual hearings.

Cybersecurity Concerns

As courts increasingly rely on technology, cybersecurity emerges as a critical concern. The survey reveals significant variation in confidence levels regarding IT security.

While 57% of respondents feel highly confident in their IT systems’ security, an alarming 22% of respondents say they are “not at all confident” in the security of their IT systems.

Generational Workforce Changes

The survey identifies generational workforce shifts as another major factor affecting courts. Baby Boomers and Gen Xers exiting the workplace, along with Gen Zers entering the workforce and Millennials moving into leadership positions, are trends frequently ranked as transformational or high impact.

These demographic changes have important implications for technology adoption. As the report notes, Gen Zers are digital natives who are very comfortable using technology and may find it easier to manage automated workflows, while they may be resistant to jobs and tasks that still rely heavily on manual tasks.

Reducing Operational Errors

The survey provides insights about task efficiency and error rates in court operations.

Entering and updating data in court management systems was rated as both the most error-prone task by a wide margin and also as the second-most inefficient task. This finding suggests that greater use of automation in CMS entry could yield major improvements in both efficiency and error rates.

The survey also found correlations between different operational challenges. Tasks that are more stressful are also correlated with causing inconvenience for court users, suggesting that addressing workflow inefficiencies could simultaneously improve both staff satisfaction and user experience.

A Critical Juncture for Courts

The survey suggests that courts face a strategic choice: embrace AI technologies that could significantly alleviate operational pressures, or risk falling further behind as staffing challenges intensify and workloads continue to grow.

“We’re facing challenges — staff don’t think they have enough time to meet their demands, and they’re working more hours to get the work done, and that’s leading to burnout,” said David Slayton, executive officer and clerk of court for the Superior Court of Los Angeles County.

“It’s incumbent on court leaders to really think about how technology can help us with this problem.”

Mike Abbott, head of Thomson Reuters Institute, underscored the urgency of the situation.

“Courts are facing an unprecedented convergence of change, driven by generative AI and generational shifts in their workforce, at the same time as they continue to deal with staff shortages, backlogs and delays,” Abbott said.

“AI literacy can empower the courts to understand both the risks and the opportunities associated with the technology, enabling them to identify the best use cases which help them focus on higher value work.”



Source link

Continue Reading

Tools & Platforms

State AI leaders gather at Princeton to consider how the technology can improve public services

Published

on


Much of the news about artificial intelligence has focused on how it will change the private sector. But all around the country, public officials are experimenting with how AI can also transform the way governments provide essential services to citizens while avoiding pitfalls.

State AI leaders, including Gov. Phil Murphy of New Jersey, gathered at Princeton University in June to discuss how AI offers ways for government to be more efficient, effective, and transparent, especially at a time when budgets are strapped and economic uncertainty has slowed down hiring.

Hosted by Princeton’s Center for Information Technology (CITP), the NJ AI Hub, the State of New Jersey, the National Governors Association, the Center for Public Sector AI, GovLab, and InnovateUS, the conference brought together more than 100 AI leaders from 25 states to share ideas and collaborate. The meeting was conducted under an agreement of confidentiality to allow participants to discuss progress and concerns openly. Quotations in this story are used by permission.

What emerged was enthusiasm about AI’s potential to reduce the time government employees spend on manual tasks and improve their ability to engage citizens, as well as concerns about how best to use public data to innovate and increase equity rather than undermine it.

The gathering is just one of the ways that CITP – which is a joint center of the Princeton School of Public and International Affairs and Princeton Engineering – is leading on AI. The center also holds policy precepts to engage policymakers in AI governance at the SPIA in DC Center, and several affiliated faculty teach courses on AI policy at Princeton SPIA.

“There’s a clear recognition of the need for thinking about public accountability and equity,” said Princeton’s Arvind Narayanan. “At the same time, I think there’s also recognition of the potential for governments if we get this right.”

At the conference, CITP Director Arvind Narayanan noted that attendees were focused on practical implementation of AI tools rather than the “polarizing conversations around AI that dominate the media.” He also explained why public-facing deployments of AI by state governments have been slower than internal ones.

“There’s a clear recognition of the need for thinking about public accountability and equity. At the same time, I think there’s also recognition of the potential for governments if we get this right,” said Narayanan, who is also a professor of computer science and co-author of “AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference.”

Speakers shared big and small ways that AI is improving government. Some noted saving an hour or two a week per employee by leveraging AI to help draft grant applications, assess legislation, or review procurement policies while ensuring oversight and accuracy. One city automated the summarization of council oral votes, a task that was previously completed by a city clerk, creating summaries of 20 years of council books in a short period of time at nearly zero cost. As a result, voters have a simpler way to access information and hold elected officials accountable.

In his remarks, Gov. Phil Murphy laid out how New Jersey is approaching the technology, including its partnership with Princeton on the NJ AI hub.

“We held hands and jumped into the AI space,” Murphy said of the state’s partnership with the University. Together with Microsoft and New Jersey-based AI company CoreWeave, the state and University launched the NJ AI Hub earlier this year to foster AI innovation. “I don’t think we’d be all in if we didn’t think that the probabilities were very high that a lot of good things could go right with AI, but I think we also have to acknowledge some of the tensions that are still playing themselves out.”

Murphy highlighted concerns about AI’s potential to empower bad actors, as well as its impact on human creativity, jobs, and equity.

“Is this going to be something that is a huge wealth generator for the few, or are we going to be able to give access to this realm to everybody,” he said.

One of the ideas attendees considered at the conference was building a public AI infrastructure that would ensure it remains an open-source technology, rather than becoming privately controlled by a few companies. Bringing AI into the public domain would also present an opportunity to build in controls and mechanisms for accountability, speakers noted. They argued that AI is foundational infrastructure, not unlike roads, bridges, and broadband.

At the end of the two-day gathering, Anne-Marie Slaughter, chief executive of the New America Foundation and former Princeton SPIA dean, reflected on the conference. She emphasized what others had said about needing to be transparent in how AI is used and ensuring that public trust in government is strengthened.

“[AI] doesn’t just transform how government does things better, faster, cheaper. It can transform what government does and, even more importantly, what government in a democracy is,” Slaughter said. “You can start to co-create and you can start to co-govern.”

Attendees pose with Governor Murphy

Posing with Gov. Phil Murphy at the conference are (left to right) Cassandra Madison of the Center for Public Sector AI, CITP Director Arvind Narayanan, New Jersey Chief AI Strategist Beth Simone Noveck, Timothy Blute of the National Governors Association and Jeffrey Oakman, senior strategic AI Hub project manager at Princeton.

 



Source link

Continue Reading

Tools & Platforms

How Trump’s megabill could slow AI progress in US

Published

on


The elimination of federal renewable energy tax credits in Trump’s One Big Beautiful Bill Act has major implications for the global AI race.

Ultimately, the shift means slowing down US progress on new energy production, which is key to winning the technology Cold War with China. There is no possible way tech companies can power the massive rollout of AI factories without solar, and now it will be that much more expensive.

But the attempt to throw a lifeline to the fossil fuel industry could be too little, too late, as detailed in this New Yorker article by Bill McKibben. The rate of solar adoption is now about a gigawatt every 15 hours. A gigawatt is the output of a typical nuclear power plant.

Solar isn’t just cheaper than fossil fuels. It’s also faster to deploy, which is crucial in the AI race. The expansion of AI data centers is creating new economic incentives for innovation in renewables, from geothermal to fusion to new battery chemistries, which can store all that new solar power. It’s a topic I expect we’ll be covering more and more here in the coming months.



Source link

Continue Reading

Trending