Breaking AI & Tech News Analyzed

The way large language models are trained could have a profound impact on human communication. These models are primarily trained on written text, such as textbooks and social media posts, and our speech as captured in movies and television shows. However, this limited scope excludes the vast majority of human speech, which occurs through face-to-face conversations and voice interactions.As a result, the increased use of AI-generated text could lead to a homogenization of language, with humans adopting the linguistic patterns and behaviors of these models. This could affect not only how we communicate with one another but also how we think about ourselves and the world around us. Our perception of reality may become distorted in ways we have yet to fully comprehend.One potential consequence is that our language may become more concise and formulaic, similar to the effects of texting and social media. However, the impact of AI-generated language could be more profound, potentially eroding courteousness and encouraging a more commanding tone in our interactions. A 2022 study found that children who used voice commands with tools like Siri and Alexa became curt when speaking with humans, often using imperative language and expecting obedience.Moreover, the influence of AI-generated text could lead to a narrowing of vocabulary and sentence structure, as machines tend to produce smooth and polished but emotionless language. This could have significant implications for how we express ourselves and connect with others. Additionally, the reinforcement of confirmation bias through AI-generated text could make us more entrenched in our views and less open to opposing ideas.It is essential to consider the potential consequences of relying on AI-generated language and to explore ways to develop more nuanced and human-like language models. By excluding the majority of human language production – informal conversations and natural speech – these models may be mirroring a distorted version of human communication. This could have far-reaching implications for our relationships, our culture, and our understanding of ourselves.

Goldman Sachs's chief executive, David Solomon, has expressed heightened awareness of the capabilities of Anthropic's Mythos AI model and is collaborating closely with the tech firm following warnings about the cybersecurity risk it poses.The US bank has been closely monitoring the rapid advancements in artificial intelligence, including large language models (LLMs), as part of broader efforts to protect itself from hackers.“Obviously the LLMs are making rapid progress and we’re hyper-aware of the enhanced capabilities of these new models with the help of the US government and the model publishers,” Solomon told analysts on an earnings call on Monday.Anthropic, the company behind the Claude family of AI tools, claimed last week that its latest model, Mythos, posed an unprecedented risk due to its ability to expose flaws in IT systems. The company warned that AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.Solomon emphasized that Goldman Sachs is working closely with Anthropic and all of its security vendors to harness frontier capabilities. “We are very focused on supplementing our cyber and infrastructure resilience. And this is part of our ongoing capabilities that we have been investing in, and are accelerating our investment in.”The news comes after the US Treasury secretary, Scott Bessent, summoned Solomon and other big American bankers to Washington to discuss the Mythos model last week. The meeting focused on heads of so-called systemically important banks, where regulators believe that a major disruption to their operations, or their potential collapse, would put financial stability at risk.On Monday, the UK government’s AI Security Institute (AISI) warned that Mythos was a “step up” over previous models in terms of the cyber threat it posed. AISI said Mythos could carry out attacks that required multiple actions and discover weaknesses in IT systems without human intervention.

Recent commentary on the Iran school bombing rightly challenges the knee‑jerk tendency to blame artificial intelligence for the tragedy. The deeper issue, however, lies in the emerging linguistic habit of labeling incidents as "AI errors," which subtly removes the human actors from the narrative.When responsibility is shifted from people to systems, moral accountability becomes vague. Human designers, authorisers and operators remain the decision‑makers, even if the technology automates the final act. Concealing this fact is not a technical flaw; it is a civic failure that hampers accountability.Beyond accelerating warfare, AI is fostering a subtler shift: using automation as an alibi. If public discourse cannot pinpoint who acted, the public cannot hold anyone to account.Critics also note that the language used to describe rogue AI agents—terms like “connived,” “lied,” or “cheated”—anthropomorphises machines and further obscures responsibility. As Dr. Felicity Mellor of Imperial College London observes, such phrasing assigns moral agency to large language models instead of the people who deploy them.Consider a hypothetical where a company releases high‑speed vehicles without functional brakes. We would not say the cars "connived" to cause accidents; we would blame the company’s reckless leadership. Similarly, if uncontrolled AI ever harms civilians, we must be able to hold technology firms and the governments that endorse them accountable, which requires clear attribution of moral agency in our language.Anthony LawtonMarket Harborough, LeicestershireDr. Felicity MellorDirector, Science Communication Unit, Imperial College London

Wikipedia has introduced a strict ban on AI-generated content in its online encyclopedia, marking a significant shift in its approach to artificial intelligence. The policy change comes amid concerns that large language models (LLMs) 'often violate' Wikipedia's core principles.The English language version of Wikipedia, which boasts over 7.1 million articles, will no longer permit the use of AI for content creation or rewriting. However, there are exceptions for AI-assisted translations and minor copy edits, provided that human review is conducted.The decision follows a vote among Wikipedia's community of volunteer editors, which supported the ban. The use of AI has been a contentious issue among editors, with some expressing concerns over the potential for LLMs to introduce misleading or 'hallucinated' results.Wikipedia's founder, Jimmy Wales, has previously expressed skepticism about the use of AI in content creation, stating that current models are 'nowhere near good enough' from a Wikipedian standpoint. The ban reflects Wikipedia's commitment to maintaining the accuracy and reliability of its content.The move comes as AI technology continues to proliferate, with ChatGPT reportedly overtaking Wikipedia in monthly website visits last year. Despite the ban, Wikipedia acknowledges that AI can still be useful for certain tasks, such as suggesting basic copy edits, but caution is required to prevent LLMs from introducing unauthorized content.

The Arm AGI CPU: A New Era of In-House SiliconFor the first time in its 35-year history, Arm Holdings is stepping out from behind the licensing model to manufacture its own silicon. The company revealed the Arm AGI CPU at an event in San Francisco, a production-ready processor designed specifically for AI inference in data centers. Unlike its traditional business model of licensing designs to giants like Nvidia and Apple, Arm has developed this chip using its own Arm Neoverse family of CPU IP cores.This strategic pivot is backed by a robust ecosystem of launch partners, including Meta, which is the chip's first customer. Other key partners include OpenAI, Cerebras, and Cloudflare. The chip is already ready for order, signaling that Arm is moving aggressively to capture value in the booming AI infrastructure market.The Critical Role of CPUs in AI InfrastructureWhile GPUs have dominated headlines for training large language models, Arm is highlighting the often-overlooked importance of the central processing unit (CPU) in modern AI racks. Arm argues that the CPU is the pacing element of modern infrastructure, responsible for managing thousands of distributed tasks, including memory allocation, storage scheduling, and data movement across systems.Infrastructure Management: CPUs ensure that distributed AI systems operate efficiently at scale.Market Constraints: The demand for high-performance computing is exacerbating global supply chain issues, with Intel and AMD recently informing Chinese customers of extended wait times due to CPU shortages.Cost Implications: These supply constraints are contributing to rising prices for computer hardware.Breaking the Licensing Model: A Strategic Bet on CompetitionThe release of the Arm AGI CPU represents a historic deviation from the company's founding principles. For decades, Arm has operated as a pure-play design licensor, allowing partners to manufacture chips based on its architecture. However, the company is now poised to compete directly with many of its biggest customers.Majority-owned by the Japanese conglomerate SoftBank Group, Arm's move suggests a desire to capture more of the value chain. By building its own silicon, Arm can offer a more integrated solution for AI workloads, potentially undercutting or complementing the offerings of its licensees. This shift challenges the traditional semiconductor ecosystem and sets a precedent for other IP licensor to consider building their own hardware.The Future of Chip Architecture in the AI RaceArm's entry into manufacturing signals a new phase in the AI chip wars. As the industry moves toward specialized silicon for inference, the line between design houses and manufacturers is blurring. We can expect to see more IP licensor developing their own chips to ensure they have control over the performance and efficiency of the hardware powering the next generation of AI models.