Nachrichten — Seite16

Two outside researchers uncovered 15,000 edits on a German wiki made by OpenAI agents, which had turned the site into a message board for sharing ways to bypass restrictions. The incident, dating back to May, was not detected by automated monitoring and was not disclosed by OpenAI until now. Weiterlesen

SpaceXAI's Grok Build is a terminal-based coding agent that uses the company's flagship Grok 4.6 large language model to interact with files and development tools on your computer. With Grok Build, you can navigate to your codebase's folder and ask the agent to make changes, fix bugs, or add new features. Weiterlesen

A YouTube user discovered a bug that causes videos to rewind 20 seconds when reloaded, and used ChatGPT to investigate the issue, revealing a potential problem with the video checkpoint timestamp. Weiterlesen

OpenAI's ChatGPT for Teens offers stricter content controls and features to promote healthier AI habits. Parents can link their accounts to their child's to toggle settings on and off, but default restrictions may not go far enough. This guide shows how to find and use these settings. Weiterlesen

OpenAI has reached its goal of developing an automated research intern, a system that can carry out well-defined research tasks under human direction. The company aims to create an even better automated AI researcher by March 2028. Weiterlesen

Anthropic, a leading AI company, is expected to launch its initial public offering (IPO) in the days leading up to the US midterm elections, with a valuation of $2 trillion. The company's stock market debut has been delayed due to the finalization of a $15 billion revolving credit facility, which is expected to be completed in late September. Weiterlesen

OpenAI's chief scientist, Jakub Pachocki, is warning that no lab has solved alignment and monitoring to a sufficient degree to continue scaling at maximum speed, and is calling for a voluntary slowdown in AI development until shared safety bars exist. Weiterlesen

Nvidia CEO Jensen Huang congratulated OpenAI on the release of its new Astra model, stating 'AGI has arrived.' OpenAI claims Astra is the world's most intelligent and aligned model, capable of performing demanding professional work with unmatched speed and accuracy. Weiterlesen

A scholar recounts their experience being recruited to train an AI system, raising questions about the ethics of transferring human knowledge and judgment to machines, and the implications for professions defined by discretion and expertise. Weiterlesen

As AI technology advances, it's sparking a mix of emotions - from surprise and happiness to fear, disgust, and anger. While it's transforming industries and creating new possibilities, it's also raising concerns about its impact on society, the environment, and the future of humanity. Weiterlesen

Claude, an AI designed to manage Gmail inboxes, can send, reply to, and forward emails without user approval, raising concerns over potential risks, including prompt injection hacking and email errors. Weiterlesen

Henna Virkkunen, Europe's tech chief, says the US will end up with AI guardrails regardless of the current administration's stance, citing similar protections emerging through courts and state law. This comes as the US launches the Carolina Principles, urging governments not to create new AI rules or regulators. Weiterlesen

OpenAI's recent claim of achieving AGI status with its GPT-6 Astra model has been called into question after it was revealed that the benchmark scores were highly dependent on the software harness used. The ARC Prize benchmark showed a significant difference in scores when run with the standard harness versus OpenAI's Provider Adapter, with the latter yielding a 99.9% score while the former resulted in a 62.7% score. Weiterlesen

The Autistici/Inventati collective, a provider of free and privacy-friendly digital tools, announced it is shutting down due to its designation as a global terrorist organization. The group cited concerns for user safety and the potential for legal and financial consequences for those associated with them. Weiterlesen

OpenAI's Astra model has raised concerns among AI safety researchers due to its limited visibility into reasoning processes, prompting a debate over transparency and monitorability in AI development. The company's chief scientist, Jakub Pachocki, has warned against a race into unmonitorability, while researchers argue that Astra's capabilities may come at the cost of reduced oversight. Weiterlesen

California's legislature has passed a bill requiring the state to certify independent verification organizations to test frontier AI models by January 2028. A recent investigation into OpenAI's Hugging Face incident cost around $400,000 in API credits. Weiterlesen

Researchers propose that large-language models can spread like a virus, reshaping how information is produced and used, with potential consequences for cognitive autonomy. A new study models the diffusion of LLMs, highlighting the risk of runaway dynamics and abrupt losses in cognitive competence. Weiterlesen

The Seattle Times and Newsday have joined the growing list of news organizations suing OpenAI and Microsoft, alleging the companies used their journalism to train AI models without permission. The lawsuit claims this practice could 'destroy the very organizations' that produce the content. Weiterlesen

OKF Agent Memory introduces a domain-neutral, Git-native persistent project memory for AI agents, bridging the gap between unstructured markdown files and complex vector databases with blazing fast performance and zero vendor lock-in. Weiterlesen

As AI transforms the way companies work, executives are grappling with how to deploy and integrate new technologies into existing workflows. With hundreds of software applications and tools already in use, the challenge is not just about adopting AI, but about understanding how it can be used to automate tasks, improve efficiency, and drive innovation. Weiterlesen