A small number of samples can poison LLMs of any size

Hacker NewsThursday, October 9, 2025 at 4:04:04 PM
NeutralTechnology
Recent research highlights that even a small number of samples can negatively impact large language models (LLMs), raising concerns about data integrity and model reliability. This finding is significant as it underscores the importance of careful data selection and management in AI development, ensuring that LLMs remain robust and trustworthy.
— Curated by the World Pulse Now AI Editorial System

Was this article worth reading? Share it

Recommended Readings
A built-in 'off switch' to stop persistent pain
PositiveTechnology
A groundbreaking development in pain management has emerged with the introduction of a built-in 'off switch' designed to alleviate persistent pain. This innovative approach not only offers hope to those suffering from chronic conditions but also represents a significant advancement in healthcare technology. By targeting the neurological pathways associated with pain, this solution could transform the lives of millions, making it a crucial topic in ongoing medical research.
Hundreds of free VPN apps are not fit for purpose - but sadly, we can't tell you which are the naughty ones
NegativeTechnology
A recent report has highlighted significant privacy risks associated with numerous free VPN applications, revealing issues like outdated code and insecure data handling practices. This matters because many users rely on these services for online security, unaware of the potential dangers lurking within these apps. The findings raise concerns about user privacy and the effectiveness of free VPNs, prompting a call for greater transparency and security in the industry.
Researchers find just 250 malicious documents can leave LLMs vulnerable to backdoors
NegativeTechnology
A recent study reveals that as few as 250 malicious documents can expose large language models (LLMs) to significant vulnerabilities, potentially allowing for backdoor attacks. This finding is crucial as it highlights the need for enhanced security measures in AI systems, especially given their increasing integration into various sectors. The implications of such vulnerabilities could be far-reaching, affecting everything from data privacy to the reliability of AI-generated content.
LLMs are mortally terrified of exceptions
NeutralTechnology
A recent discussion highlights the challenges that large language models (LLMs) face when dealing with exceptions in data processing. This is significant because understanding these limitations can help developers improve AI systems, making them more robust and reliable in real-world applications.
Can your mouse spy on you? Research shows a high-sensitivity gaming mouse can pick up your speech with a malicious AI-driven trick
NegativeTechnology
Recent research has raised concerns about the potential for high-sensitivity gaming mice to be used as surveillance devices through AI technology. While it's unlikely that your mouse is actually spying on you, the findings highlight a growing vulnerability in our everyday devices. This matters because it underscores the need for better security measures in technology we often take for granted.
'Bring your own device is the norm': BYOD has reached epidemic levels as more workers use their personal smartphone for work - and businesses scramble to catch up
NegativeTechnology
The rise of Bring Your Own Device (BYOD) policies has become a significant concern for businesses as more employees use personal smartphones and laptops for work. Recent research highlights that this trend is leaving companies exposed to security vulnerabilities, prompting a scramble to adapt and implement better management strategies. It's crucial for organizations to address these challenges to protect sensitive information and maintain productivity.
What MIT got wrong about AI agents: New G2 data shows they’re already driving enterprise ROI
PositiveTechnology
Recent data from G2 challenges MIT's claims about AI projects, revealing that 95% are not failing as previously suggested. In fact, nearly 60% of companies have successfully deployed AI agents, with a failure rate of less than 2%. This highlights the growing effectiveness of AI in driving enterprise ROI, contradicting earlier academic forecasts of stagnation in AI development. Understanding this shift is crucial for businesses looking to leverage AI technology for competitive advantage.
Has Your Scientific Work Been Cut? We Want to Hear.
NegativeTechnology
The Times is launching a new series to highlight the impact of policy changes by the Trump administration on scientific research. They are seeking stories from scientists whose work has been cut, shedding light on the broader implications for innovation and knowledge in the scientific community. This matters because it underscores the challenges faced by researchers and the potential loss of valuable contributions to society.
Energy Firms Snap Up Weather Services for Trading Edge in Japan
PositiveTechnology
In Japan's rapidly evolving power-trading market, weather forecasters are capitalizing on the demand for specialized data, providing energy firms with crucial insights to navigate one of the world's most volatile electricity markets. This trend not only highlights the importance of accurate weather predictions in energy trading but also signifies a growing intersection between technology and traditional industries, ultimately benefiting both forecasters and energy companies.
How to clear your Windows 11 PC cache (and what happens if you don't)
PositiveTechnology
Clearing the cache on your Windows 11 PC can significantly enhance its performance, making it run faster and more efficiently. This article provides a straightforward guide on how to locate the cache settings and clear out unnecessary data. Regularly managing your cache is essential because neglecting it can lead to sluggish performance and potential issues with your system's responsiveness.
European Union Outlines Strategies to Boost AI Adoption, Research
PositiveTechnology
The European Union has unveiled new strategies aimed at enhancing the adoption and research of artificial intelligence. This initiative is crucial as it addresses growing concerns that Europe is falling behind in the tech race against China and the U.S. By prioritizing AI, the EU aims to foster innovation and ensure that it remains competitive in the global market, which could lead to significant advancements in various sectors.
Study of 1M-year-old skull points to earlier origins of modern humans
PositiveTechnology
A recent study of a 1 million-year-old skull has provided new insights into the origins of modern humans, suggesting that our species may have emerged earlier than previously thought. This discovery is significant as it challenges existing theories and enhances our understanding of human evolution, potentially reshaping how we view our ancestry and the development of early human traits.
Latest from Technology
The Xiaomi 15T Pro may borrow from the competition, but I love it anyway
PositiveTechnology
The Xiaomi 15T Pro is making waves in the smartphone market, drawing comparisons to the iPhone with its sleek design and advanced features. This model stands out for its impressive performance and user-friendly interface, making it a strong contender in a competitive landscape. Fans of Xiaomi will appreciate the brand's ability to innovate while still appealing to those who love the aesthetics of the iPhone. It's exciting to see how this device could influence future smartphone designs.
Tsunami threat widens to parts of Indonesia and Palau after earthquake off southern Philippines
NegativeTechnology
A tsunami threat has expanded to parts of Indonesia and Palau following a significant earthquake off the southern coast of the Philippines. This situation is concerning as it poses risks to coastal communities, highlighting the need for preparedness and timely warnings in the region. The potential for a tsunami can lead to devastating impacts, making it crucial for residents and authorities to stay alert and take necessary precautions.
Senate Passes AI Chip Export Limits on Nvidia, AMD to China
NegativeTechnology
The Senate has passed legislation imposing export limits on AI chips from Nvidia and AMD to China, requiring these companies to prioritize US access to their products. This move is seen as a significant setback for the tech industry, which has been actively trying to block such measures. The implications of this legislation could affect global supply chains and the competitive landscape in the AI sector, highlighting ongoing tensions between the US and China.
Crypto Operator HashKey Said to Mull Hong Kong Listing This Year
PositiveTechnology
HashKey Group, known for operating Hong Kong's largest licensed crypto exchange, is reportedly considering an initial public offering (IPO) this year. This move could significantly impact the local crypto market, attracting more investors and enhancing the region's reputation as a financial hub. The potential listing reflects growing confidence in the cryptocurrency sector and could pave the way for further developments in the industry.
7.4-magnitude earthquake strikes offshore from southern Philippines
NegativeTechnology
A powerful 7.4-magnitude earthquake struck offshore from a southern province in the Philippines early Friday morning, raising concerns about a potential tsunami in the area. This significant seismic event highlights the ongoing geological risks faced by the region, reminding residents and authorities of the importance of preparedness and response strategies in the face of natural disasters.
Thailand’s Delta Shares Set to Drop as Exchange Imposes Curbs
NegativeTechnology
Shares of Delta Electronics (Thailand) Pcl, the nation's leading publicly traded company, are expected to decline after the stock exchange introduced market surveillance measures due to a recent record-breaking rally. This is significant as it reflects the regulatory body's response to unusual market activity, which could impact investor confidence and the overall market dynamics.