The Answer Engine AI News · Sydney

Twenty-five prominent mathematicians have signed an open letter accusing AI labs of threatening their intellectual work. The dispute centres on how AI companies like OpenAI use mathematical research to train their models. Mathematicians argue their published work is being ingested into AI systems without proper attribution, compensation, or regard for how it might affect the integrity of mathematical knowledge. This follows a broader pattern of tension between AI developers and content creators across multiple fields—writers, artists, and now academics are pushing back against how their work is used to build commercial AI products. The open letter represents a significant escalation, bringing together leading figures in mathematics to formally challenge the practices of major AI labs. While the specific demands of the letter aren't detailed in the source, the conflict highlights growing concerns about intellectual property, fair use, and the relationship between AI companies and the knowledge producers whose work powers their systems.

Chinese AI company Moonshot AI, the maker of the Kimi chatbot, is aiming for $2 billion in annual revenue. The company's K3 models are seeing significant usage, with OpenRouter data showing up to 300 billion tokens being generated daily by K3 models on their platform. However, the figures tell a slightly mixed story: while token generation remains high, K3's overall usage has experienced a modest decline in recent months. Moonshot AI is one of China's leading AI startups, and Kimi has positioned itself as a competitor to ChatGPT and other Western chatbots in the Chinese market. The $2 billion revenue target signals the company's confidence in monetising its large user base and maintaining its position in China's competitive AI landscape. The high token generation numbers suggest Kimi is still being used heavily for longer, more complex tasks that require substantial processing power, even if raw user numbers may have plateaued.

Staff working inside leading AI laboratories are publicly raising concerns that advanced artificial intelligence systems could pose an existential threat to humanity. MIT Technology Review gathered its AI editorial team to examine whether these warnings from industry insiders reflect genuine danger or amount to scaremongering. The discussion features executive editor Niall Firth, senior AI editor Will Douglas Heaven, and AI reporter Grace Huckins unpacking the so-called "AI extinction" scenario. The debate is significant because it's coming from people building these systems, not just outside critics. However, the conversation also questions whether apocalyptic framing serves the AI industry's interests by attracting attention and investment, or whether it distracts from more immediate harms like bias, misinformation, and job displacement. For business owners trying to make sense of AI's trajectory, this represents the core tension in today's AI discourse: separating legitimate long-term concerns from hype, while staying focused on practical, near-term impacts that affect operations, customers, and strategy today.

Researchers have developed a mathematical framework that models how groups of people make decisions together, treating collective behaviour like interconnected probability waves rather than independent choices. Testing this on Chinese stock market trading data, they found that 82-94% of trader decisions followed "adaptive entangled" patterns – meaning traders' choices were interconnected and influenced each other in real-time – rather than acting as independent rational actors (which occurred less than 5% of the time). Only 2-12% of behaviours were direct responses to news or events. The research indirectly supports a hypothesis about "nonlocal entangled nerve fibres" in the brain, suggesting our neural networks may be more interconnected than previously understood. The framework aims to capture human intelligence behaviours through analytical mechanisms that could apply beyond financial markets. While this is theoretical research published on arXiv (not yet peer-reviewed), it challenges fundamental assumptions about independent decision-making that underpin both economic models and many AI systems designed to predict or influence human behaviour.

Researchers have compared two ways AI agents can use reusable knowledge libraries to tackle complex, multi-step tasks. The traditional approach loads "skill packages" – bundles of instructions, scripts and resources – directly into an AI agent's working memory, and the agent follows those instructions. The problem: as tasks get longer and more complex, the agent's context window fills up with information, and its reasoning quality degrades. The alternative tested is "subagent execution," where each skill package spawns a fresh AI agent in a clean context window to handle that specific subtask. The study found subagents outperform the traditional approach when skill packages have clear inputs and outputs, and contain detailed procedural instructions. The downside is extra communication overhead between the main agent and subagents. This matters because as businesses deploy AI agents for increasingly complex workflows, understanding how to structure them for reliability becomes critical.

SEO For Paws is running another free online conference featuring well-known search industry figures including Barry Schwartz and Glenn Gabe. The event focuses on search in the AI era and brings together search experts to discuss the latest developments. It's a charity initiative, meaning the conference supports a good cause while delivering current search marketing insights. The live stream format means Australian business owners can tune in without travel costs or time zone hassles typically associated with international conferences. While the source summary doesn't provide specific session details or dates, these types of industry events typically cover practical topics like how AI-powered search engines are changing visibility strategies, what businesses need to adapt, and real-world case studies. The "big names" draw suggests quality speakers who work directly with search platforms and see data across many industries. Because it's free and live-streamed, it's accessible to small businesses who normally couldn't justify conference ticket prices or international travel.

A Search Engine Journal article revisits a content strategy from 2003 that unknowingly anticipated how modern AI search systems work. The framework focused on two core tactics: using nested phrases (variations of keywords within your content) and publishing content when it's timely and relevant. These approaches were designed to help websites rank for the wide variety of actual searches people type into Google, rather than just optimising for one or two exact keywords. What's notable now is that this same approach aligns perfectly with how AI-powered search engines and answer engines operate today. These systems generate longer, more conversational queries and need to match content against a broader range of question formats. The article suggests that businesses don't necessarily need entirely new strategies for AI search – the fundamentals of writing naturally, covering topic variations, and publishing when your audience is actively searching remain effective. It's a reminder that good content practices often translate across technology shifts, and that "query fan-out" – reaching multiple related searches with one piece of content – has been a sound approach for over two decades.

A new study from the University of Washington has found that Google's AI Overviews feature reduced search referrals to Wikipedia by approximately 5%. The research provides early evidence of how AI-generated answers at the top of search results may be affecting click-through rates to traditional web sources. Google has disputed the findings, though the company hasn't released its own data to counter the claim. This matters because Wikipedia is one of the most-linked sources on the web and a common reference point in search results—if AI Overviews are reducing traffic even to a site as dominant as Wikipedia, smaller websites and businesses could face similar or greater impacts. The 5% drop may sound modest, but for businesses relying on search traffic for leads and sales, even small percentage decreases can translate to meaningful revenue losses. The study adds to growing evidence that Google's shift toward AI-generated answers is changing how users interact with search results, potentially keeping them on Google rather than clicking through to source websites.

iTnews hosted a roundtable lunch discussion at Society restaurant in Melbourne focusing on security challenges posed by "shadow AI" — the use of AI tools and services within organisations without IT department oversight or approval. The roundtable brought together industry professionals to discuss this emerging issue. While the source material consists primarily of photographs from the event rather than detailed discussion points, the topic itself reflects growing concern among Australian businesses about employees using consumer AI tools like ChatGPT, Claude, or Copilot for work purposes without proper security protocols, data governance, or compliance measures in place. This creates potential risks around data leakage, intellectual property protection, and regulatory compliance that many organisations are only beginning to address.