News

Chinese cyberespionage campaign renews efforts in multiple organizations in Southeast Asia, blending tactics and expanding efforts 

New research evaluating the effectiveness of reward modeling during Reinforcement Learning from Human Feedback (RLHF): “SEAL: Systematic Error Analysis for Value ALignment.” The paper introduces quantitative metrics for evaluating the effectiveness of modeling and aligning human values:

Abstract: Reinforcement Learning from Human Feedback (RLHF) aims to align language models (LMs) with human values by training reward models (RMs) on binary preferences and using these RMs to fine-tune the base LMs. Despite its importance, the internal mechanisms of RLHF remain poorly understood. This paper introduces new metrics to evaluate the effectiveness of modeling and aligning human values, namely feature imprint, alignment resistance and alignment robustness. We categorize alignment datasets into target features (desired values) and spoiler features (undesired concepts). By regressing RM scores against these features, we quantify the extent to which RMs reward them ­ a metric we term feature imprint. We define alignment resistance as the proportion of the preference dataset where RMs fail to match human preferences, and we assess alignment robustness by analyzing RM responses to perturbed inputs. Our experiments, utilizing open-source components like the Anthropic preference dataset and OpenAssistant RMs, reveal significant imprints of target features and a notable sensitivity to spoiler features. We observed a 26% incidence of alignment resistance in portions of the dataset where LM-labelers disagreed with human preferences. Furthermore, we find that misalignment often arises from ambiguous entries within the alignment dataset. These findings underscore the importance of scrutinizing both RMs and alignment datasets for a deeper understanding of value alignment.

As technology rapidly advances, the boundaries of what’s possible in personal computing are continuously expanding. One of the most exciting innovations on the horizon is the concept of the AI PC, which stands for Artificial Intelligence Personal Computer. AI PCs accounted for 14% of all personal computers shipped in the second quarter of 2024, with demand expected to continue to grow.

These intelligent machines are set to transform the way we interact with our computers, offering unprecedented performance and personalization. Let’s delve into what an AI PC is, explore the benefits it offers consumers, and understand how it is reshaping the future of computing.

What is an AI PC

An AI PC is a computing device that integrates artificial intelligence capabilities directly into its hardware and software. Unlike traditional PCs, which rely on external software or cloud services for AI functionalities, AI PCs have built-in AI processors or coprocessors that enable them to perform intelligent tasks locally.

These machines leverage advanced AI algorithms to enhance various aspects of computing, from performance and efficiency to user experience and security. They have a neural processing unit (NPU), “a type of processor designed to handle the mathematical computations specific to machine learning algorithms.” NPU speed is now measured by “trillions of operations per second” (TOPS).

Benefits of AI PCs

By embedding AI capabilities into the core of the PC, these devices can offer a more responsive, personalized, and secure computing environment. Here’s how they are transforming personal computing:

1. Enhanced Performance Through Intelligent Automation

One of the standout features of AI PCs is their ability to automate and optimize tasks intelligently. AI PCs can learn from user behavior and system performance to streamline processes and improve efficiency. For example, AI can manage system resources dynamically, prioritizing tasks based on current needs and usage patterns. This means that applications requiring high performance, such as gaming or video editing, can run more smoothly without manual intervention.

AI algorithms can also predict and pre-load applications and files that users are likely to access next, reducing load times and improving overall responsiveness. This level of automation and optimization ensures that users experience a seamless and efficient computing environment.

2. Improved Efficiency in Processing Data-Intensive Applications

Data-intensive applications, such as those used for machine learning, scientific research, and complex simulations, benefit greatly from the power of AI PCs. These machines are equipped with specialized AI processors designed to handle large volumes of data quickly and efficiently. By offloading specific tasks to these AI processors, the main CPU is freed up to handle other operations, resulting in faster processing speeds and reduced latency.

For professionals and researchers working with big data or computationally heavy applications, AI PCs can drastically cut down processing times and enhance productivity. The integration of AI ensures that these applications can perform complex calculations and analyses with greater accuracy and speed.

3. Personalized User Experiences

AI PCs excel in delivering personalized user experiences by learning and adapting to individual preferences and behaviors. Through continuous learning, AI systems can customize the operating environment based on how users interact with their PCs. This can include adjusting system settings, recommending software or files, and even optimizing user interfaces to align with personal habits and preferences.

For example, an AI PC might analyze your work patterns and suggest tools or shortcuts that enhance productivity. It can also personalize your entertainment experience by recommending media content based on your viewing history and preferences. This level of personalization creates a more intuitive and enjoyable user experience.

4. AI-Driven Threat Detection and Mitigation

Cybersecurity has become a constant underlying threat in the digital age. Last year, 880,418 Americans reported cybercrime to the FBI’s Internet Crime Complaint Center, which was a 10% increase from 2022.

AI PCs are addressing this issue with advanced threat detection and mitigation capabilities. AI-driven security systems can analyze patterns and behaviors to identify potential threats such as malware, phishing attempts, or unauthorized access. AI-driven security systems use machine learning algorithms to detect threats in real-time. This proactive approach enhances the protection of sensitive data and ensures a safer computing environment.

5. Assistance with Everyday Personal Tasks

AI PCs are not just about high-performance computing and security; they also excel in assisting with everyday personal tasks. For instance, AI-powered virtual assistants integrated into the PC can help manage schedules, set reminders, and perform routine tasks such as composing emails or creating documents.

These virtual assistants learn from user interactions to offer more accurate and contextually relevant assistance. They can also automate repetitive tasks, such as file organization or data entry, saving users time and effort. By handling mundane activities, AI PCs allow consumers to focus on more complex and creative tasks.

The Future of AI PCs

The integration of AI into personal computing is a glimpse into the future of technology. As AI PCs become more advanced, we can expect even greater enhancements in performance, efficiency, and user experience. These devices are not just about adding new features; they represent a fundamental shift in how we interact with technology, making computing more intuitive, personalized, and secure.

As we move forward, keeping an eye on these advancements will be crucial in harnessing their full potential and embracing the next era of personal computing. The future of AI PCs is here, and it’s poised to redefine how we interact with our digital world.

The post What is an AI PC? appeared first on McAfee Blog.

Highline Public Schools in Washington State have now been closed for two days following the incident

TIDRONE group targets military, drone and satellite industries in Taiwan

The Justice Department has begun the latest round of fraud reimbursement from the Western Union Remission Fund

Researchers have warned that a critical SonicWall vulnerability is being exploited in ransomware attacks

Seit Langem hält sich der Glaube, dass das macOS-Betriebssystem weniger anfällig für Schadsoftware ist als Windows. Das mag an der geringeren Marktdominanz liegen und verschiedenen eigenen Sicherheitsfeatures, die von den Malware-Entwicklern andere Ansätze verlangen. Man ging davon aus, dass hier nur unkonventionelle Attacken und Schadsoftware eine Chance hätten. Diese Annahme ist nun endlich passé. Mainstream-Schadsoftware […]

According to Microsoft researchers, North Korean hackers have been using a Chrome zero-day exploit to steal cryptocurrency.

Tom Hanks, one of the most recognizable faces in the world, warns that scammers have swiped his likeness in malicious AI deepfakes.

As reported by NBC News, Actor Tom Hanks issued an announcement to his followers saying his name, likeness, and voice have shown up in deepfaked ads that promote “miracle cures” without his consent. The actor posted on Instagram:

In the ever-evolving landscape of digital advertising, a new challenge has emerged that blurs the lines between reality and artificial fabrication: AI-generated content using celebrity likenesses.

Tom Hanks isn’t the only victim. Earlier in 2024, we saw a malicious AI deepfake of Taylor Swift front a phishing scam with a free cookware offer. In 2023, the deepfaked likeness of Kelly Clarkson pushed weight loss gummies. And, just a few weeks ago, malicious deepfakes of Prince William endorsed a bogus investment platform. We’ve also seen deepfakes of noteworthy researchers hawking miracle cures as well, which we’ll soon cover in another blog post.

Without question, we live in a time where scammers can turn practically anyone into a deepfake. The AI tools used to create them have only gotten better, more accessible, and easier to use. Compounding that concern is just how convincing these bogus endorsements look and sound.

The cost of celebrity deepfakes

Malicious deepfakes affect more than the celebrities they mimic. They affect everyone who goes online. As we’ve seen with Tom Hanks, while deepfakes can potentially tarnish his reputation, they can also harm the general public. By pushing disinformation and frauds, deepfakes open the door to health risks, identity theft, and in an election year, voter suppression — as we saw with the Joe Biden AI voice clone robocalls in Vermont.

Celebrities like Scarlett Johansson have begun to fight back legally against the unauthorized use of their likenesses. However, the legal framework in the U.S. remains largely unprepared for the challenges posed by AI-generated content. Yet we’re seeing some progress, at least on a state level in the U.S.

Tennessee recently issued a piece of legislation that says state residents have a property right to their own likeness and voice. In effect, Tennesseans can take legal action if another person or group creates deepfakes in their likeness. Illinois and South Carolina have similar legislation under consideration.

Those represent just a handful of 151 state-level bills that have been introduced or passed through July of this year — all covering AI deepfakes and deceptive media online.[i] Likewise, we’ll take a closer look at how legislation is catching up with AI in an upcoming blog.

Using AI to fight AI deepfakes

As we’re quick to point out in our blogs, not all AI deepfakes are bad. AI deepfake tools have plenty of positive uses, such as dubbing and subtitling movies, creating training and “how-to” videos, and even creating harmless and humorous parody videos — all well within the scope of the law.

The problem is with malicious deepfakes, like the ones Tom Hanks warned us about. Yet how can you spot them?

Technology has kept pace, as it has with our newly released Deepfake Detector. It alerts you in seconds if it spots AI-manipulated content. Right in your browser. It works like this:

Deepfake Detector monitors audio being played through your browser while you browse. If it determines what you’re watching or listening to contains AI-generated audio, it alerts you right away.

McAfee doesn’t store any of this audio or browsing history. What you watch is yours, and you get to keep that private.

It works in the background while you browse. So, if a deepfake Tom Hanks or Taylor Swift video crops up in your feed, you’ll know with a high degree of confidence that it’s a fake. You can easily snooze notifications or turn off scanning right from your dashboard.

Deepfake Detector shows how much is real and how much is fake. With a browser extension, Deepfake Detector shows what portion of audio was deepfaked, and at what point in the video that content cropped up. Think of it working like a lie detector in the movies. As the video plays, peaks of red lines and troughs of gray lines show you what’s likely a fake and what’s likely real.

Our new role in the face of AI and malicious deepfakes

As AI-detection technology continues to advance, the responsibility also falls on us, collectively, to keep an eye out for fakes. Especially the glut of malicious deepfakes we now face.

The key to navigating this new era of AI is awareness. Indeed, tools will help us spot deepfakes. Yet we can count on ourselves to spot them too.

First off, we need to realize just how easy it is to create a deepfake. Keeping that in mind keeps us on guard. Next, when we see that celebrity gushing about a miracle cure or another promoting a screaming great deal, we know to stop and think before we act.

From there, we have plenty of excellent and reputable fact-checking resources that can help us get to the truth. Snopes, Reuters, Politifact, the Associated Press, and FactCheck.org all offer great ways to find out if what we’re seeing and hearing is true, false, or somewhere in between.

And with this kind of awareness in mind, we’ve launched the McAfee Smart AI Hub. We see the rise of malicious deepfakes as a major concern. It’s a security concern. An identity theft concern. A health concern. An election concern. And a family concern as well. We created the hub with these in mind and established it as a place where you can learn about the latest AI threats. Additionally, it’s a place where you can join the fight against malicious deepfakes by turning in the ones you find online.

While the advent of AI brings remarkable benefits, it also introduces complex challenges. As we move forward, balancing innovation with ethical considerations and consumer protection will be paramount. Without a doubt, we’ll continue to follow it all closely here in our blogs.

As for the Tom Hanks deepfakes, if something seems too good to be true, like miracle advice, it probably is. Stay curious, stay cautious.

[i] https://www.brennancenter.org/our-work/research-reports/states-take-lead-regulating-ai-elections-within-limits

The post Tom Hanks Warns Fans: The Dark Side of AI Scams  appeared first on McAfee Blog.