Firefox 135 debuts today, with (what else?) AI built right into the browser. Unlike Microsoft Edge and Copilot or Brave's own Leo AI chatbot, Firefox isn't investing in its own AI capabilities. Instead, the company is building in AI access to third-party chatbots such as Anthropic's Claude and ChatGPT, plus lesser-known LLMs like HuggingChat and Le Chat Mistral. As of a week ago, this AI capability was labeled "experimental" and was being pushed to a small part of the Firefox user base. Now, the Firefox release notes state that it's being pushed out gradually to all users.

DeepSeek has said that its most recent model was trained on Nvidia H800s. This is an A.I. chip that Nvidia developed specifically for the Chinese market after export controls were first imposed, and that caused a fair amount of drama in Washington. When the United States put restrictions on Nvidia's most advanced chips in 2022, Nvidia quickly adapted by creating slightly downgraded chips that fell just under the threshold the government had set. These chips were technically legal for Chinese companies to use, but allowed them to achieve practically the same results. This angered Biden officials, and they moved to restrict the new chips as well. But the government moved slowly, and it took them about a year to ban the H800 and other downgraded chips.

American stock markets shuddered on Monday, prompted by China's announcement that it has created a capable, cheap, artificial intelligence machine. It's the biggest cloud yet to darken the West's blue-sky enthusiasm over AI, calling into question the efficacy of America's export controls and the billions of dollars the United States is pouring into the technology's expensive cutting edge. China startup DeepSeek says its AI assistant uses less advanced chips than its rivals' models do, and it costs less to train. Unlike the West's billions, the Chinese model was developed for just 5.6 million, by one estimate. "Are we going to spend 500 billion to get to the frontier so that China can find a way to copy our homework for pennies on the dollar?"

The tech sector's ongoing effort to force-feed generative AI features into widely used services continues with updates to WhatsApp's ChatGPT integration. There are now more ways to interact with the chatbot in the app. To use it there, you have to add 1 (800) CHAT-GPT -- or 1 (800) 242-8478 -- to your contacts. WhatsApp says users around the world can now upload images to the chatbot, just like they can when using ChatGPT directly. You can also send a voice message to talk to ChatGPT, which will respond using text.

An elite group of companies known as the "Six Tigers"--Stepfun, Zhipu, Minimax, Moonshot, 01.AI, and Baichuan--are generally considered to be at the forefront of China's AI sector. But alongside them, research-focused firms like DeepSeek and ModelBest continue to grow in influence. Some, such as Minimax and Moonshot, are giving up on costly foundational model training to hone in on building consumer-facing applications on top of others' models. Others, like Stepfun and Infinigence AI, are doubling down on research, driven in part by US semiconductor restrictions. We have identified these four Chinese AI companies as the ones to watch.

Everything to do with artificial intelligence has been the big IT hype of the past two years. Even if the initial enthusiasm for ChatGPT and others has now given way to a more sober assessment, there is hardly a software company at the moment that is not taking a close look at the possibilities of the technology. Microsoft in particular has invested huge sums in AI development and is demonstrating how AI can also be integrated into familiar programs: Gradually, more and more applications are being given functions that fulfill their tasks with the help of artificial intelligence. Microsoft has also released its Large Language Model (LLM) Copilot as its own app and browser extension. Other companies have now also embedded AI functions into apps, some of which are available for free.

Within AI, though, what impact is DeepSeek likely to have in the longer term? Here are three seeds DeepSeek has planted that will grow even as the initial hype fades. First, it's forcing a debate about how much energy AI models should be allowed to use up in pursuit of better answers. You may have heard (including from me) that DeepSeek is energy efficient. That's true for its training phase, but for inference, which is when you actually ask the model something and it produces an answer, it's complicated.

OpenAI's CEO Sam Altman has signed a deal with South Korean tech giant Kakao as his firm seeks new alliances after Chinese rival DeepSeek shook the global AI industry. Altman says the partnership, which lets Kakao use ChatGPT for its artificial intelligence services, will prioritise safety.

OpenAI said on Tuesday it will develop artificial intelligence products for South Korea with chat app operator Kakao, unveiling a second major alliance with a high-profile Asian partner this week. In a whirlwind tour through Asia, OpenAI CEO Sam Altman also announced a partnership with Japan's SoftBank on Monday and is, according to sources, scheduled to visit India on Wednesday where he is seeking to meet with Prime Minister Narendra Modi. Like SoftBank, Kakao said it would be using technology developed by the ChatGPT creator for its products.

Large Language Models (LLM)-based systems, i.e. interconnected elements that include an LLM as a central component (e.g., conversational agents), are typically monolithic static architectures that rely on a single LLM for all user queries. However, they often require different preprocessing strategies, levels of reasoning, or knowledge. Generalist LLMs (e.g. GPT-4) trained on very large multi-topic corpora can perform well in a variety of tasks. They require significant financial, energy, and hardware resources that may not be justified for basic tasks. This implies potentially investing in unnecessary costs for a given query. To overcome this problem, a routing mechanism routes user queries to the most suitable components, such as smaller LLMs or experts in specific topics. This approach may improve response quality while minimising costs. Routing can be expanded to other components of the conversational agent architecture, such as the selection of optimal embedding strategies. This paper explores key considerations for integrating routing into LLM-based systems, focusing on resource management, cost definition, and strategy selection. Our main contributions include a formalisation of the problem, a novel taxonomy of existing approaches emphasising relevance and resource efficiency, and a comparative analysis of these strategies in relation to industry practices. Finally, we identify critical challenges and directions for future research.