The Clop ransomware gang is targeting Internet-exposed Gladinet CentreStack file servers in a new data theft extortion campaign. […] Read More
Law enforcement has seized the servers and domains of the E-Note cryptocurrency exchange, allegedly used by cybercriminal groups to launder more than $70 million. […] Read More
(c) SANS Internet Storm Center. https://isc.sans.edu Creative Commons Attribution-Noncommercial 3.0 United States License. Read More
The North Korean threat actor known as Kimsuky has been linked to a new campaign that distributes a new variant of Android malware called DocSwap via QR codes hosted on phishing sites mimicking Seoul-based logistics firm CJ Logistics (formerly CJ Korea Express). “The threat actor leveraged QR codes and notification pop-ups to lure victims into […]
RePo: Language Models with Context Re-Positioningcs.AI updates on arXiv.org arXiv:2512.14391v1 Announce Type: cross
Abstract: In-context learning is fundamental to modern Large Language Models (LLMs); however, prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional indices. Drawing on Cognitive Load Theory (CLT), we argue that this uninformative structure increases extraneous cognitive load, consuming finite working memory capacity that should be allocated to deep reasoning and attention allocation. To address this, we propose RePo, a novel mechanism that reduces extraneous load via context re-positioning. Unlike standard approaches, RePo utilizes a differentiable module, $f_phi$, to assign token positions that capture contextual dependencies, rather than replying on pre-defined integer range. By continually pre-training on the OLMo-2 1B backbone, we demonstrate that RePo significantly enhances performance on tasks involving noisy contexts, structured data, and longer context length, while maintaining competitive performance on general short-context tasks. Detailed analysis reveals that RePo successfully allocate higher attention to distant but relevant information, assign positions in dense and non-linear space, and capture the intrinsic structure of the input context. Our code is available at https://github.com/SakanaAI/repo.
arXiv:2512.14391v1 Announce Type: cross
Abstract: In-context learning is fundamental to modern Large Language Models (LLMs); however, prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional indices. Drawing on Cognitive Load Theory (CLT), we argue that this uninformative structure increases extraneous cognitive load, consuming finite working memory capacity that should be allocated to deep reasoning and attention allocation. To address this, we propose RePo, a novel mechanism that reduces extraneous load via context re-positioning. Unlike standard approaches, RePo utilizes a differentiable module, $f_phi$, to assign token positions that capture contextual dependencies, rather than replying on pre-defined integer range. By continually pre-training on the OLMo-2 1B backbone, we demonstrate that RePo significantly enhances performance on tasks involving noisy contexts, structured data, and longer context length, while maintaining competitive performance on general short-context tasks. Detailed analysis reveals that RePo successfully allocate higher attention to distant but relevant information, assign positions in dense and non-linear space, and capture the intrinsic structure of the input context. Our code is available at https://github.com/SakanaAI/repo. Read More
Within the past year, artificial intelligence copilots and agents have quietly permeated the SaaS applications businesses use every day. Tools like Zoom, Slack, Microsoft 365, Salesforce, and ServiceNow now come with built-in AI assistants or agent-like features. Virtually every major SaaS vendor has rushed to embed AI into their offerings. The result is an explosion […]
Since the end of the year is quickly approaching, it is undoubtedly a good time to look back at what the past twelve months have brought to us… And given that the entire cyber security profession is about protecting various systems from “bad things” (and we’ve all correspondingly seen more than our share of the […]
Cisco has alerted users to a maximum-severity zero-day flaw in Cisco AsyncOS software that has been actively exploited by a China-nexus advanced persistent threat (APT) actor codenamed UAT-9686 in attacks targeting Cisco Secure Email Gateway and Cisco Secure Email and Web Manager. The networking equipment major said it became aware of the intrusion campaign on […]
Microsoft has confirmed that recent Windows updates trigger RemoteApp connection failures on Windows 11 24H2/25H2 and Windows Server 2025 devices in Azure Virtual Desktop environments. […] Read More
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Wednesday added a critical flaw impacting ASUS Live Update to its Known Exploited Vulnerabilities (KEV) catalog, citing evidence of active exploitation. The vulnerability, tracked as CVE-2025-59374 (CVSS score: 9.3), has been described as an “embedded malicious code vulnerability” introduced by means of a supply chain compromise Read […]