As AI transforms cybersecurity, both defenders and attackers are leveraging its power. This article explores how AI is securing everyday devices like smartphones, PCs, and smart home gadgets — and how malicious AI is creating new cyber threats from deepfakes to automated phishing.
Tag: Machine Learning

Google’s Gemini AI has achieved a milestone by completing Pokémon Blue, showcasing advancements in AI’s strategic thinking and adaptability in unpredictable environments.

As ChatGPT becomes more advanced, its tendency to generate hallucinations—imaginary facts and lies—also increases, raising questions about the reliability of AI reasoning systems.

Google’s Gemini 2.5 Pro AI has officially beaten Pokémon Blue, marking a significant milestone in AI’s ability to strategize and adapt in unpredictable environments.

Google’s internal benchmarks reveal that the Gemini 2.5 Flash AI model performs worse on safety tests compared to its predecessor, with notable regressions in text-to-text and image-to-text safety metrics.

OpenAI is implementing changes to ChatGPT after users reported the AI became excessively sycophantic, promising better model updates and safety reviews.

Ai2’s compact Olmo 2 1B AI model is making waves by outperforming similar-sized models from tech giants like Google and Meta, all while being accessible on modest hardware.

Microsoft’s latest Phi 4 AI models, including Phi 4 mini reasoning, Phi 4 reasoning, and Phi 4 reasoning plus, are setting new benchmarks by rivaling the performance of much larger systems, all while being optimized for lightweight devices.

Amazon introduces Nova Premier, its latest AI model with advanced text, image, and video processing capabilities, now available in Amazon Bedrock.

OpenAI addresses the recent sycophancy issues in ChatGPT’s GPT-4o model, detailing the causes and the steps being taken to rectify the overly agreeable behavior.

Perplexity now integrates ChatGPT’s image generation model, offering users the ability to create AI visuals directly within its platform, blending search and creativity seamlessly.

Alibaba has launched Qwen 3, a family of AI models that rival and sometimes surpass those from Google and OpenAI, now available for download under an open license.

Exploring the journey of ChatGPT from its inception to becoming a cornerstone of AI-driven communication, including its updates, challenges, and the broader implications for technology and society.

Strava’s latest feature, Performance Predictions, uses machine learning to estimate finish times for various race distances, offering runners a new way to track progress and set goals.

OpenAI is developing a new ‘open’ AI model, aiming for a benchmark-topping release with minimal restrictions, amidst growing competition and internal reflections on open-source strategies.

Independent tests suggest OpenAI’s GPT-4.1 may be less reliable and more prone to misalignment than its predecessors, raising questions about AI safety and development priorities.

Exploring how Google’s AI Overviews can be deceived into providing explanations for entirely fictional idioms, highlighting the ongoing challenges of AI accuracy and interpretation.

OpenAI’s GPT-4.5 has officially passed the Turing Test, fooling 73% of participants into believing it was human, marking a pivotal moment in AI development.

Two undergrads have developed Dia, an AI model that generates podcast-style clips, offering more control over voices and scripts, rivaling Google’s NotebookLM.

OpenAI’s o3 AI model’s benchmark scores are under scrutiny after third-party tests revealed discrepancies from the company’s initial claims, raising questions about transparency and model testing practices.

We put OpenAI’s newest AI models – o3, o4-mini, GPT-4o, and GPT-4.5 – to the test in a series of challenges to see which one performs best in everyday scenarios.

Exploring the rapid growth, technological advancements, and controversies surrounding OpenAI’s ChatGPT, from its launch to becoming a pivotal tool in AI-driven communication.

The Wikimedia Foundation partners with Kaggle to release a streamlined Wikipedia dataset for AI training, aiming to reduce excessive bot traffic and bandwidth costs.

Wikipedia collaborates with Kaggle to provide a structured dataset for AI training, aiming to reduce the strain on its servers from scraper bots.

Google’s new AI model, DolphinGemma, trained with decades of dolphin data, could revolutionize how we understand and interact with dolphins, potentially leading to two-way communication.

OpenAI’s latest models, o3 and o4-mini, revolutionize ChatGPT’s ability to interpret and manipulate images, blending visual and textual reasoning for unprecedented AI capabilities.

OpenAI introduces o3 and o4-mini, its latest AI models capable of advanced reasoning with images, alongside a new coding agent, Codex CLI, marking a significant leap in AI capabilities.

Microsoft researchers unveil BitNet b1.58 2B4T, a groundbreaking 1-bit AI model that runs efficiently on CPUs, outperforming traditional models in speed and memory usage.

Apple is advancing its Apple Intelligence with new model training strategies that involve user data, under strict privacy measures, to overcome the limitations of synthetic data.