Tag: DeepSeek

DeepSeek, the Hangzhou‑based artificial‑intelligence firm, announced the release of two new reasoning‑capable models, V3.2 and V3.2‑Speciale. V3.2 is now available through the company’s app and web platform, while V3.2‑Speciale is offered via API access only. The company claims the Speciale version surpasses Google’s Gemini 3.0 Pro and OpenAI’s GPT‑5 High on benchmark tests, positioning DeepSeek as a serious contender in the global AI race. Leggi di più

Researchers from the Institute of Strategic Dialogue examined four widely used AI chatbots and found that they frequently reference Russian state‑affiliated media and other sanctioned sources when answering questions about the war in Ukraine. The study highlights how data gaps can be exploited by disinformation networks and raises concerns about the ability of large language models to filter prohibited content, especially within the European Union. Leggi di più

Microsoft Edge now displays a "Try Copilot" button when users visit popular AI services such as ChatGPT, Perplexity and Deepseek. The button opens the Copilot assistant in the browser sidebar, a move Microsoft says is meant to promote its own AI offering. The prompt does not appear on visits to Google’s Gemini, and a larger pop‑up appears when users navigate to the Chrome download page, urging them to stay with Edge. Some users find the nudges intrusive and have discovered how to disable the Copilot button through the browser’s settings. Leggi di più

ByteDance’s Doubao app has overtaken DeepSeek to become the most popular AI chatbot in China, drawing tens of millions of monthly active users. Designed with a friendly avatar and deep integration with Douyin, Doubao offers text, audio, video, image generation and customizable agents. Analysts credit ByteDance’s expertise in viral app design and seamless cross‑platform sharing for Doubao’s rapid growth, while DeepSeek remains a more minimalist text‑only service. The company is now exploring extensions of Doubao into smart glasses, cars and other devices, signalling an ambition to broaden the AI assistant beyond smartphones. Leggi di più

Assembly language, the low‑level code that speaks directly to a processor, has shaped both iconic video games and cutting‑edge artificial‑intelligence research. Chris Sawyer used x86 assembly to build the beloved titles RollerCoaster Tycoon and Transport Tycoon, trading modern tools for raw efficiency and personal passion. Decades later, companies such as DeepSeek and DeepMind have revisited assembly to squeeze performance out of Nvidia chips, demonstrating that even today’s AI breakthroughs can hinge on the meticulous control that only assembly provides. The story illustrates how a language once deemed archaic remains vital in today’s tech landscape. Leggi di più

DeepSeek, a Chinese AI lab, has vaulted into mainstream awareness as its chatbot app climbs to the top of major app stores. Backed by High‑Flyer Capital Management, the company has released a series of models—including DeepSeek‑V2, DeepSeek‑V3, and the reasoning‑focused R1—that claim strong benchmark performance and low inference costs. The rapid rise has drawn attention from industry giants, regulators, and governments, prompting both integration into platforms like Microsoft’s Azure AI Foundry and bans on government devices in several jurisdictions. Leggi di più

DeepSeek announced a new experimental AI model featuring Sparse Attention technology that dramatically lowers inference costs for long‑context tasks. The model, released on Hugging Face and accompanied by a research paper on GitHub, uses a lightning indexer and fine‑grained token selection to focus computational resources on the most relevant excerpts. Preliminary tests suggest API call prices can be cut by as much as 50 percent in long‑context scenarios. The open‑weight release invites third‑party validation and positions DeepSeek as a notable player in the ongoing effort to make transformer‑based AI more cost‑effective. Leggi di più

DeepSeek is testing a sparse attention technique aimed at cutting the processing costs of large AI language models. By limiting the number of word‑to‑word comparisons, the approach seeks to mitigate the quadratic scaling problem inherent in traditional transformer architectures. The effort could make long‑form interactions more affordable while maintaining the model’s ability to understand context. Leggi di più

DeepSeek, a Chinese AI lab, has rapidly risen in popularity after its chatbot app topped major app store charts. The company’s models, including DeepSeek‑V2, V3 and the R1 reasoning model, have demonstrated strong benchmark performance while being priced far below competitors. This success has drawn attention from U.S. The firm benefits from backing by High‑Flyer Capital Management and has partnered with Microsoft’s Azure AI Foundry, yet it faces bans on government devices in the United States and other jurisdictions due to security and propaganda concerns. Leggi di più

DeepSeek announced its experimental model V3.2‑exp, featuring a new Sparse Attention mechanism that dramatically lowers inference expenses for long‑context tasks. The architecture employs a lightning indexer to prioritize excerpts and a fine‑grained token selector to feed a limited attention window, allowing the model to process extensive context with reduced server load. Preliminary tests suggest API calls in long‑context scenarios could cost up to half as much as before. The model is open‑weight and freely available on Hugging Face, inviting independent verification and broader adoption. Leggi di più

Knowledge distillation, a method that transfers information from a large "teacher" model to a smaller "student" model, has become a fundamental tool for reducing the size and expense of AI systems. Originating from a 2015 Google paper, the technique leverages soft‑target probabilities to convey nuanced relationships between data classes, enabling compact models to retain high performance. Over the years, distillation has been applied to language models such as BERT and its distilled variant, DistilBERT, and is now offered as a service by major cloud providers. Recent developments continue to expand its utility across reasoning tasks and open‑source initiatives. Leggi di più