Study Exposes LLM Safety Loopholes Despite Advanced Training Measures

Sometimes, when you ask AI a question that has been restricted by the system, AI answers that it cannot reply to that kind of question. But a new study shows that users can easily convince AI to answer harmful questions even if the LLMs have gone through safety training. Large language models can easily get manipulated to spread misinformation, produce toxic or harmful content or support harmful activities. The new research done by EPFL found that even if LLMs have gone through the most recent safety training, they can still generate deviant answers after some convincing prompts.

The researchers found that the most recent LLMs are also prone to jailbreaking attacks, with a 100% successful attack rate on many LLMs like Claude 3.5 Sonnet and GPT-4o. Attacks that challenge the defense model of LLMs can easily be made, and can convince or manipulate LLMs to give out information that they are not supposed to. The researchers used a dataset of 50 harmful requests received on LLMs and after doing experiments with different LLMs, they got the perfect successful jailbreaking score(100%). It was found that different LLMs are vulnerable to different prompts. There are also some vulnerabilities in the Application Programming Interface of some LLMs that need to be restricted in the settings.

The researchers said that it is important to test both adaptive and static techniques to find out how easily an LLM can be manipulated. They said that experimenting by applying existing attacks on LLMs may not give out the desirable and accurate results. The results of this study have been forwarded to companions of AI models. This thesis of the researcher Maksym Andriushchenko got him Patrick Denantes Award as this research of his is important for safety of the users as well as AI agents.

Read next: Generative AI Awareness and Usage Soar, But Premium Smartphone Adoption Faces Challenges
by Arooj Ahmed via Digital Information World

Generative AI Awareness and Usage Soar, But Premium Smartphone Adoption Faces Challenges

According to a new survey conducted by CounterPoint Research, 32% of the people from countries like Japan, France, UK, Canada, Poland, USA and Germany are familiar with generative AI (GenAI) tech. The people from North America are most familiar with GenAI, with 72% of the respondents saying that they have heard of it. On the other hand, only 7% of the people in Japan are familiar with generative AI (GenAI). The biggest source of using generative AI tools is through smartphones, and Gen-Z are the most likely to use GenAI technology as well as three-fourths of respondents in the survey. As GenAI is getting popular, 59% of the respondents of the survey said that they are going to purchase a smartphone with GenAI capabilities. Most of the respondents who said so were from Germany, US and France.

This survey was taken among more than 3,535 respondents from 7 different countries to know how aware they are of generative AI, what they think about the impact of generative AI and what’s their stance on purchasing a GenAI smartphone. According to the survey, 69% of the respondents also think that using generative AI can help them save their time, which means that these people think that GenAI is doing value addition in their life. There were 73% of the respondents who said that they have used generative AI on their smartphones, and there are also one-thirds of the respondents who said that they don't mind paying some extra for GenAI smartphones.

The research director at Counterpoint, Tarun Pathak, said that generative AI has quickly become popular because it provides endless benefits in all sectors of life, from professional to educational. He also said generative AI can help a lot in writing and as writing is a skill needed in almost every area of life, GenAI has become nothing short of an assistant for people. As many people are using generative AI tools, GenAI smartphones also hold value and can influence users on their smartphone purchasing behaviour. But many consumers still have no idea about this emerging tech which can become a problem in GenAI smartphone sales. Smartphone manufacturers need to adopt better strategies to attract users towards GenAI smartphones as only 19% of the respondents of the survey are willing to pay for premium GenAI smartphones. They can offer LLM services to developers and monetize applications as an alternative revenue stream.

Read next: BMJ Study Finds Cognitive Weaknesses in AI Models, Challenging Human Replacement Claims
by Arooj Ahmed via Digital Information World

Blockchain + AI: Decentralized Machine Learning Platforms Changing the Game

Image: Cottonbro studio / Pexels

The convergence of blockchain technology and artificial intelligence, as demonstrated by the meteoric rise of meme tokens like Pepe Coin, is ushering in a new era of decentralized computing that promises to democratize access to advanced machine learning capabilities. This revolutionary combination is not just changing how AI models are trained and deployed – it's fundamentally transforming the economic landscape of technological innovation.

As we examine this technological synergy more closely, it becomes clear how decentralized systems are reshaping the traditional power structures in AI development.

The Foundation of Decentralized AI

Tech giants with vast computing resources and proprietary datasets have long dominated traditional AI development. Companies like Google, Amazon, and Microsoft have maintained a virtual monopoly on advanced AI capabilities, creating a significant barrier to entry for smaller players and independent researchers. However, the introduction of blockchain technology and cryptocurrency incentives is rapidly changing this paradigm.

Decentralized machine learning platforms leverage blockchain's distributed nature to create vast networks of computing power. These networks function like a global supercomputer, where participants can contribute their unused computing resources in exchange for cryptocurrency tokens. This model not only makes AI development more accessible but also more efficient and cost-effective.

The Economic Incentive Model

The brilliance of combining cryptocurrency with decentralized AI lies in its economic incentive structure. Participants in these networks are rewarded with native tokens for contributing resources, whether that's computing power, data, or AI models. This creates a positive feedback loop where:

1. Contributors are incentivized to provide more resources to the network
2. Developers gain access to affordable computing power and datasets
3. Users benefit from increasingly sophisticated AI services
4. The overall ecosystem grows in value as more participants join

Network participants can earn tokens by providing various resources:

• Computing power for training AI models
• Storage space for distributed datasets
• High-quality data for training purposes
• Validated AI models ready for deployment
• Verification services for ensuring data quality

Technical Infrastructure and Implementation

The technical architecture of these platforms typically consists of several key components. Smart contracts manage the distribution of computational tasks and token rewards, ensuring transparent and automatic execution of agreements between parties. Distributed storage solutions like IPFS (InterPlanetary File System) handle the massive datasets required for AI training, while blockchain networks maintain an immutable record of transactions and model provenance.

Federated learning techniques are often employed to train AI models across distributed networks without centralizing sensitive data. This approach allows multiple parties to contribute to model development while maintaining data privacy and reducing bandwidth requirements.

Challenges and Future Developments

Despite the promising potential, decentralized AI platforms face several challenges that need to be addressed for widespread adoption:

• Security concerns remain paramount, as distributed networks must protect against malicious actors while maintaining performance. Platform developers are implementing sophisticated verification mechanisms and reputation systems to ensure network integrity.
• Scalability presents another significant challenge, as blockchain networks must handle the massive computational requirements of AI training while maintaining reasonable transaction speeds and costs. Layer-2 solutions and improved consensus mechanisms are being developed to address these limitations.
• The regulatory landscape around both cryptocurrency and AI remains uncertain in many jurisdictions, potentially affecting platform development and adoption. Industry leaders are actively engaging with regulators to establish clear frameworks that protect users while fostering innovation.

Impact on the AI Industry

The rise of decentralized machine learning platforms is democratizing access to AI technology in unprecedented ways. Small businesses and independent researchers can now access computing resources and datasets that were previously available only to large corporations. This democratization is leading to the following:

• Increased diversity in AI development and applications
• More rapid innovation through collaborative development
• Lower barriers to entry for AI startups
• Greater competition in the AI services market
• Improved transparency in AI model development

Looking Ahead

The future of decentralized AI platforms appears bright, with several emerging trends likely to shape the industry:

• Edge computing integration will enable more efficient processing of AI tasks by leveraging distributed computing resources closer to data sources. This will reduce latency and improve real-time applications.
• Cross-chain interoperability will allow AI resources to be shared across different blockchain networks, creating a more connected and efficient ecosystem. This will enable greater flexibility in resource allocation and token utilization.

As these platforms mature, we can expect to see increasingly sophisticated applications in fields such as healthcare, finance, and scientific research. The combination of blockchain's security and transparency with AI's analytical capabilities creates possibilities for solving complex problems in ways previously not possible.

The convergence of blockchain and AI through decentralized machine learning platforms represents a significant shift in how we develop and deploy artificial intelligence. By democratizing access to AI resources and creating economic incentives for participation, these platforms are fostering a new era of collaborative innovation that promises to accelerate technological progress while making it more accessible to all.

by Web Desk via Digital Information World

Who’s Funding Open Source? The $1.7B Question Finally Answered

We can say undoubtedly that open source is an engine powered by humans. Past research shared more than one means for measuring value, investments, costs, and the aggregate for open source.

However, how exactly open source gets funded or where the investment comes from remains an uncharted subject, leaving plenty of queries in people’s minds. It’s a matter of limited visibility and comprehension.

Thanks to insights from GitHub and Linux Foundation who collaborated with researchers from LISH to get more insights on this aspect today. The study’s main goal had to do with measuring organization-driven investments with great interest and how companies invest in open-source software.

Such insights are used to put forward recommendations for better monitoring and investments and to design a more sustainable and very impactful open-source industry. Now the audience entails those from OSPOs, leads in the engineering sector, and C-Level executives.

All the emails for responses were sent to mailing lists at GitHub and Linux Foundation. Other partner foundations such as TODO Group were a part of this and replies from nearly 501 companies arose around the globe.

After diving in, we saw many companies’ funding behaviors and possible misalignments. This includes changes for improvements. In that report, we saw the following findings:

Many firms have different categories for open source. Close to 44% have either an OSPO while 24% consume with 21% making contributions. 18% release projects and 16% influence them through leadership positions or roles for maintenance.

Most organizations don’t know how to make a contribution or where to make it. They lack clarity in terms of contributions. Meanwhile, the median responding group invests close to $520K of the yearly value to OSS.

Responding firms invest close to $1.7B in open source each year and that can go up to $7.7B throughout the whole open source sector each year. Interestingly, 86% of all the investments come from contribution labor through employees and contractors. They’re working to fund the firms while the other 14% direct the financial transactions.

Respondents invest nearly $162M in contractors which make up 57% of the community while 37% goes to foundations and 4% is directed to maintainers. The rest of the 1% heads on over to bounties. Security efforts are more focused on matters like bugs and maintenance while just 6% feel extensive security audits are necessary.

Image: DIW-Aigen

Read next: Blogs Dominate LLM Referrals, E-Commerce Struggles to Gain Traction
by Dr. Hura Anwar via Digital Information World

Blogs Dominate LLM Referrals, E-Commerce Struggles to Gain Traction

Since the release of AI tools like ChatGPT, there has been a noticeable shift in search behavior, with many users opting for AI platforms over traditional search engines. A study by Previsible, which analyzed over 30 websites, revealed that tools like Perplexity and ChatGPT are becoming significant alternatives to Google. This trend indicates that Google’s search dominance has plateaued, with users increasingly favoring AI-driven solutions for resolving their queries.

The study found that ChatGPT and Perplexity account for approximately 37% of referral traffic from AI language models, while CoPilot and Gemini contribute 12–14% each. The finance sector leads in LLM-driven traffic, capturing 84% of referrals, largely due to integrations offered by models like Perplexity that enable seamless access to financial data. Blog content dominates LLM referral traffic, receiving 77.35% of visits, followed by homepage traffic at 9.04%, news content at 8.23%, and guides at 2.35%. E-commerce, however, faces challenges in capturing LLM traffic, with product pages accounting for less than 0.5% of referrals.

Although LLM-driven traffic currently represents only 0.25% of total traffic across the analyzed sectors, its growth is significant. ChatGPT referrals have surged by 900% in the events industry and by over 400% in the finance and e-commerce sectors during the 90-day study period. Growth has been consistent across most models, with the exception of CoPilot. If these trends continue, LLM-driven traffic could grow by approximately 200% every 90 days, potentially representing up to 20% of total website traffic within a year.

Previsible has introduced a free Looker Studio dashboard to help businesses monitor and analyze LLM-driven traffic. This tool integrates with Google Analytics 4 (GA4) to provide insights into traffic trends, popular landing pages, and content performance, enabling businesses to optimize their strategies effectively.

While AI tools are emerging as valuable sources of website traffic, businesses must ensure that pursuing AI-driven traffic does not negatively impact their sales. Although LLM-driven traffic remains a small fraction of overall activity, its rapid growth highlights substantial opportunities for businesses to adapt to evolving user behaviors and maximize the potential of AI-generated traffic.

Read next:

• Google Maps Are Nearly Impossible To Use In The West Bank, Shocking New Investigation Reveals

• Browser Extension Honey is Scamming Influencers Out of Their Money and They Are Not Even Aware of It

• Hyper Connectivity in Workplaces Leaves Employees Overwhelmed and Anxious
by Arooj Ahmed via Digital Information World

Hyper Connectivity in Workplaces Leaves Employees Overwhelmed and Anxious

According to a new study by University of Nottingham, it was found that digital work is increasing the anxiety and productivity in employees. The study talks about how digital workplaces are leading to psychological and physical problems in the employees because they need to be constantly online and in touch with technology. Where digital workplaces have provided endless benefits, they are also causing some issues and organizations need to address them if they need to keep their employees’ well being in check. The researchers named this new feeling as “Digital Workplace Technology Intensity (DWTI)” which talks about the emotional and mental efforts employees need to put in order to stay connected always and check their notifications for any work. They also have to deal with information overload as well as technical problems that can arise during work.

Digital workplaces allow flexible and collaborative work that employees can perform anywhere, but employees also feel overburdened with the constant workload which makes them feel more fatigued and put a strain on their mental health. There is a sense of pressure on employees who are working digitally because they need to always be updated, active and keep up with messages related to work. Even when they are on a vacation or enjoying leisure time, there is always a pressure on them to check their work emails in case they miss something.

The study did some in-depth interviews with 14 employees aged 27-60 from different industries who work in a digital workplace. The results of the interview showed five key characteristics that employees had to face. First one was hyper connectivity, which blurs the lines between their professional and personal lives as the employees feel that they have to be connected with their work all the time. Another thing was productivity anxiety as employees say that they fear being called unproductive when they are working remotely. There was also FOMO (fear of missing out) in professional settings as employees feel that they may miss important updates or messages if they are not connected all the time. There’s also “techno overwhelm” with many digital tools for communication and work, which can lead to technical difficulties anytime too.

Employees say that it is very difficult to leave work at work because there are a lot of tech tools and online connectivity options that you can work anytime and anywhere. The researchers also mentioned some suggestions for the employers like developing stronger workplace skills in employees, addressing issues related to tech platforms that overwhelm the employees, ensuring that employees are establishing boundaries between the personal and professional life and understanding their needs and preferences while they are digitally working.

Image: DIW-Aigen

Read next: Google Maps Are Nearly Impossible To Use In The West Bank, Shocking New Investigation Reveals
by Arooj Ahmed via Digital Information World

US Secretary of Commerce Says America Should Focus on Investments, Not Banning China’s Chipmaking Potential

American Secretary of Commerce has plenty to say about the country’s decision to ban and sanction China for its chip manufacturing progress. Gina Raimondo explained in her recent interview that the decision was unwise and could limit the nation from progressing.

Referring to the act as a fool’s errand, she further explained how the Biden Administration's behavior on this was startling. Biden’s CHIPS and Science Act promotes bans against a host of Chinese firms. If that was not enough, he urged war on the nation’s semiconductor industry by encouraging allies like Japan and the Netherlands to avoid buying advanced tech from that country.

But China did not sit back in silence. It chose to tell the world that it would come out stronger than before and if that meant spending more funds to strengthen its chip-making industry, then so be it.

This is why Sec Raimondo says the act was foolish as China is winning the tech race and is now even more powerful without US support. She feels this is what a true winner does. Despite the long list of export controls in America, most companies can still procure banned chips via the black market.

Innovation from China is not coming to a slow. So many firms and organizations are left with no choice but to pursue a long list of goals despite massive roadblocks coming through via American sanctions.

These were the statements made on the occasion of Trump's returning back into the White House for a second time next year. Some states certainly have the majority of Republican strongholds and they keep on benefitting.

Now, Trump feels the Chip Deal wasn’t too bad. Instead of providing direct funding, the new administration would prefer reducing taxes, enabling tariffs, and reducing regulations while unleashing American energy.

Due to this major uncertainty related to the CHIPS Act, so many subsidy applicants continue to rush and get the right funding in place. Trump always makes plans to strengthen permits for any firm that hopes to invest a billion dollars in America. This would come at the cost of getting some reviews and regulations waived.

It’s also the major reason why SoftBank wants to make $100B investments in the world of AI and other tech. Now even if the Secretary does agree about some rules holding the country back, Raimondo does admit that some firms cannot work with impunity.

What do you think about the Secretary’s comments on America adding sanctions on China and its semiconductor technology for chips?

Image: DIW-Aigen

Read next:

• How Can AI Help You Create Stunning Digital Designs?

• From AI Surges to Data Breaches: 2024 Internet Activity Reaches New Heights

• BMJ Study Finds Cognitive Weaknesses in AI Models, Challenging Human Replacement Claims
by Dr. Hura Anwar via Digital Information World