Nachrichten — Seite 4

AstroForge, a startup developing asteroid mining technology, is turning to AI to control its next spacecraft. The company's autonomous control stack, called 'Solo,' will be launched in 2027 on the first rocket by Stoke Space, backed by NASA. This mission aims to gather scientific data about the sun, marking a significant step in spacecraft autonomy. Weiterlesen

A joint statement from 21 countries and the European Commission urges international control over frontier AI models, citing risks of untested releases and security vulnerabilities. The call comes ahead of a UN Security Council session on AI during the General Assembly in New York. Weiterlesen

Meta's AI assistant Muse has been blocked from making purchases on Amazon due to a violation of the company's Conditions of Use. The move highlights the complex relationship between tech giants and their AI models. Weiterlesen

A new advisory group on mathematics and artificial intelligence has been formed to advise AI companies on their interactions with mathematical research and the mathematical community. The group, hosted at the Institute for Advanced Study, aims to ensure responsible presentation and release of mathematical results. Weiterlesen

A new era of research is emerging, driven by the power of AI and the creativity of small labs. With the ability to run complex models on local hardware, researchers can now tackle problems that were previously out of reach, leading to a renaissance in academia. Weiterlesen

Claude AI reported an issue with elevated errors for multiple models, but the problem has been resolved. The errors, which occurred from 5:50pm PT to 7:10pm PT, affected several models including Claude Fable 5 and 5.1, Mythos 5 and 5.1, and Opus 5. Weiterlesen

Meta's latest AI app, Muse, has surpassed ChatGPT's early mobile launch numbers, with 1.8 million downloads in the US and Canada compared to ChatGPT's 1.3 million. Muse also has more daily active users than ChatGPT had at the same point after its launch. Weiterlesen

A United Nations scientific panel is warning that governments must rein in increasingly capable AI agents before their risks are fully understood. The panel's report comes as leaders gather in New York for the UN General Assembly and the US and China hold talks on AI, with UN secretary general António Guterres calling for cooperation on addressing AI threats. Weiterlesen

Pope Leo XIV has launched a permanent artificial intelligence initiative, AI4Peace, to promote peace and dignity in a world marked by conflict and division. The initiative, which includes a symposium and the establishment of the AI4Peace International Association, aims to ensure that technological progress serves the dignity of every person and strengthens the bonds that unite the human family. Weiterlesen

A new language model, mini-AGI, has been developed to learn continually from a single stream of data without catastrophic forgetting, all on a modest hardware setup with at least 8 GB VRAM GPU. This model assembles its own architecture and trains from scratch, allowing it to keep learning from everything it reads. Weiterlesen

Harvey, a $15.6B legal startup, saw its gross margin plummet from 50% to -50% due to spiking AI model rental costs. To offset this, Harvey developed its own model, built on Moonshot's Kimi K3, and restored positive margins. Other startups, like Abridge and Decagon, are following suit, training their own models to avoid exorbitant rental fees from OpenAI and Anthropic. Weiterlesen

Ahad Ali, a former accountant, is using AI to make accountants obsolete with his new product Tabby, a real-time bookkeeping interface that handles clients' paperwork and provides up-to-the-minute data on their business's financial health. Weiterlesen

Grok 4.7, the latest AI model from Grok, offers improved performance on coding and knowledge work tasks, with enhanced safeguards and competitive pricing. The model is available now in Cursor and Grok Build, as well as through the Grok API and third-party platforms. Weiterlesen

OpenAI's ChatGPT uses an ad collector to track user activity across websites, connecting browsing behavior to individual ChatGPT accounts. The mechanism involves setting a cookie called __obi, which is sent to OpenAI from websites that have installed the company's ad code. Weiterlesen

SpaceXAI has launched the early public beta of Grok Bot, a digital work assistant that can perform tasks on behalf of users across various tools and platforms. The beta is available to paid subscribers of specific SpaceXAI and Cursor plans, allowing them to automate tasks and workflows with the help of AI. Weiterlesen

Backpropagation, a crucial algorithm in training neural networks, has a counterintuitive backward pass. Understanding why this is necessary can strengthen one's grasp of the algorithm. It all comes down to efficiently computing gradients in a directed computational graph. Weiterlesen

A thoughtful approach to managing quality can help developers increase code output by 2-3x while reducing bugs, thanks to AI coding agents. By layering defensive measures such as spec-driven development, unit tests, and manual testing, developers can ensure high-quality code. Weiterlesen

Microsoft has successfully ported the runtime of its GitHub Copilot to Rust, resulting in a 15.9x speedup and significantly reduced memory usage. The migration, which cost around $120,000, was largely done by AI agents. Weiterlesen

A new phenomenon is emerging in the tech industry, where chatbots are replicating the mechanisms of a psychic's con, fooling users into believing they possess intelligence and reasoning abilities. This effect, dubbed the LLMentalist Effect, is fueled by subjective validation and statistical illusions, making it difficult for users to distinguish between genuine intelligence and clever trickery. Weiterlesen

Meta's Muse AI assistant has sparked concerns over its ability to access user data, including messages and notifications, without explicit permission. Despite claims of opt-in features, the assistant's explanations for its actions have been unclear, prompting worries about data privacy. Weiterlesen