Tech Unplugged

Baidu has launched its updated AI models, ERNIE 4.5 and ERNIE X1 , with ERNIE Bot now accessible for free to individual users . The new models boast enhanced multimodal capabilities and strong reasoning , with ERNIE X1 achieving comparable performance to DeepSeek's R1 at a lower cost. These advancements will be integrated across Baidu's various platforms, and API access is available for enterprise...

The provided text constitutes excerpts from a comprehensive cybersecurity report for 2024-2025, outlining prevalent trends, significant threats exemplified by recent breaches, and crucial best practices. The report details escalating cybercrime costs, the emergence of AI-powered attacks, and the persistent danger of ransomware. It further examines cloud security challenges, the impact of regulator...

Facebook engineers Avinash Lakshman and Prashant Malik developed Cassandra to address the scaling challenges of Inbox Search. This distributed storage system manages structured data across numerous commodity servers, ensuring high availability and no single point of failure . Cassandra employs consistent hashing and gossip protocols for data distribution and fault detection. Its data model is base...

Cassandra is a decentralized, structured storage system initially developed by Facebook for its Inbox Search feature and later open-sourced. The document outlines Cassandra's architecture , including its data model (tables with column families), partitioning via consistent hashing, and replication strategies for high availability. It details the system's implementation , covering persistence compo...

This paper introduces "H-CoT," a novel method to bypass safety mechanisms in large reasoning models (LRMs) like OpenAI's models, DeepSeek-R1, and Gemini 2.0 Flash Thinking. By manipulating the model's chain-of-thought reasoning, the attack disguises harmful requests within educational prompts, highlighted by the new "Malicious-Educator" benchmark. Experiments show that H-CoT significantly reduces...

This research investigates the security vulnerabilities of large language models (LLMs) used for translating natural language into SQL queries (Text-to-SQL) , specifically focusing on the threat of backdoor attacks . The authors introduce ToxicSQL, a novel framework to create stealthy backdoors that can lead to the generation of malicious, yet executable, SQL queries through semantic and character...

SubLimetechie's comprehensive report analyzes numerous global large language models (LLMs) , contrasting their backend architectures , reliability measures , scalability , training data , and performance . The analysis spans proprietary models like GPT-4 and Claude 3.5 to open-source options such as Llama 3.2 and Falcon 180B , also including regional leaders. The report helps understand the streng...

This podcast from Tech Unplugged, dated March 2025, explores the collaboration between browser APIs and Web Application Firewalls (WAFs) in order to better protect against client-side attacks . It defines WAFs and common client-side threats like XSS, CSRF, and clickjacking . The report then outlines various browser APIs relevant to security and details WAF features designed for client-side defense...

A security researcher at Harel Security Research discovered a vulnerability allowing for ChatGPT account takeover. This was achieved through a "wildcard" web cache deception. The researcher found that a broad caching rule for the /share/ path, combined with a difference in how Cloudflare's CDN and OpenAI's web server parsed URL-encoded path traversals, led to the flaw. By crafting a specific URL,...

This compilation of excerpts from various sources explores the multifaceted landscape of artificial intelligence, examining its ethical, legal, policy, and societal implications. Contributions from legal scholars, philosophers, and technology experts analyze topics such as the explainability and regulation of AI, its impact on fundamental rights and media, and the challenges it poses to establishe...

This podcast provides a root cause and thesis analysis of CVE-2025-0108, an authentication bypass vulnerability in Palo Alto Networks' PAN-OS software. The flaw allows unauthenticated attackers to execute specific PHP scripts due to discrepancies in how different web technologies handle requests. Exploitation details, similar past vulnerabilities like those in Fortinet FortiOS and Apache Struts, a...

In this advanced edition podcast episode, Senior Security Researcher delves into the intricate landscape of Active Directory security. The discussion covers sophisticated attack methodologies, practical lab exercises, and essential defensive techniques. Emphasizing a red team approach, listeners will gain insights into privilege escalation, trust abuse, and mitigation strategies, while understandi...

This podcast outlines potential security weaknesses within 5G networks. It highlights vulnerabilities such as trust exploitation between devices and base stations and weak initial message protection. The introduction of new 5G features like network slicing also brings increased risk by creating more points of potential attack. Implementation and configuration errors can also expose networks, exace...

A significant data breach at Mars Hydro, a manufacturer of IoT grow lights, exposed 2.7 billion records due to inadequate security measures. The unprotected database contained sensitive user information such as Wi-Fi passwords and IP addresses. Security flaws included a lack of encryption, password protection, and regular security audits. This breach highlights critical vulnerabilities within the...

In this episode, we delve into the concept of web session management, exploring how session identifiers and cookies operate to facilitate user interactions. We highlight the threats posed by session hijacking, including techniques like man-in-the-middle attacks, session fixation, and sidejacking. Tune in to learn practical measures to safeguard your online activities, such as avoiding public Wi-Fi...

This podcast explains the SystemBC Remote Access Trojan (RAT), which has evolved to target Linux systems, posing a significant threat to corporate networks, cloud servers, and IoT devices. It highlights the malware's technical capabilities, such as its encrypted communication, lateral movement, and persistence mechanisms, as well as common attack vectors like unpatched vulnerabilities and compromi...

The Gemma 3 Technical Report introduces Google DeepMind's Gemma 3, a new generation of lightweight open-source language models. These models offer enhanced capabilities including multimodal understanding, longer context windows (up to 128K tokens), and improved multilingual abilities. The report details architectural improvements focused on memory efficiency and training methodologies involving kn...

Docker's architecture and its role in containerization. It details the shared kernel approach that distinguishes containers from virtual machines. The piece also examines resource isolation, container runtimes, networking, storage, and security aspects within the Docker ecosystem. Furthermore, the article discusses container resource architecture, focusing on namespaces and control groups for proc...

A critical review of Dan Sullivan's Google Cloud Certified Professional Cloud Architect Study Guide examining its strengths and weaknesses as a learning tool for aspiring cloud architects

CAMEL is an open-source framework designed to study the scaling laws of AI agents by simulating large-scale multi-agent systems. It provides tools and resources for researchers to experiment with different agent types, tasks, and environments. The framework emphasizes evolvability, scalability, and statefulness , using code as prompts for agent behavior. CAMEL supports various applications , inclu...

The Agent Communication Protocol (ACP) is a framework for standardizing communication between autonomous agents. It aims to enable seamless interaction, collaboration, and integration across diverse agent-based systems. ACP leverages JSON-RPC and supports both stateful and stateless transport mechanisms like REST and WebSockets. Key considerations include balancing state management, ensuring scala...

A technology that facilitates interaction between AI agents and various online services. The speaker highlights mCP's potential to revolutionize how users interact with the internet, envisioning a shift from browser-based interactions to agent-driven tasks. mCP acts as a "USBC connector" allowing agents to access and utilize diverse applications through a standardized protocol. The transcript also...

Active Directory (AD) security, a critical aspect of enterprise cybersecurity. It highlights AD's role in identity and access management, making it a prime target for attacks. The guide explores fundamental concepts, attack vectors like credential theft and privilege escalation, and reconnaissance techniques used by adversaries. Furthermore, the guide provides practical strategies for planning and...

Large Language Models (LLMs) through a question-and-answer format, covering fundamental concepts and advanced techniques. It explains tokenization, LoRA/QLoRA, beam search, and temperature, along with masked language modeling and sequence-to-sequence models. The text further explores model training methodologies, including autoregressive versus masked models, embeddings, next sentence prediction,...

AWS Security Specialty certification emphasises hands-on experience with AWS security services, secure internet protocols, and the ability to make informed trade-off decisions regarding cost and security. The material covers incident response planning, risk assessment, logging, monitoring, and infrastructure security. The course highlights IAM, key management, and data protection methods, both at...