All articles tagged with #safeguards
Anthropic is trialing a feature that lets Claude execute tasks directly on a user’s computer after prompts from a phone, enabling actions like opening apps, navigating browsers, and exporting PDFs with permission prompts; the move adds to the push for autonomous AI agents competing with Nvidia-backed OpenClaw, and integrates with Dispatch for ongoing task management.
Anthropic says it will not comply with Pentagon demands on AI safeguards as a looming deadline approaches, signaling a confrontation over how the company’s AI tech should be regulated for military use.
ByteDance pledged to strengthen safeguards for its Seedance 2.0 AI video generator after cease‑and‑desist letters from Disney and Paramount accusing the tool of unauthorized IP use and likenesses; Hollywood groups and unions criticized the practice. Seedance 2.0, launched Feb. 12, creates 15‑second clips from prompts and has sparked both praise for realism and controversy over creating copyrighted characters. ByteDance says it respects IP rights and is taking steps to prevent infringement, including pausing uploads of real people’s images as it works to address industry concerns.
The IAEA's Director General Rafael Grossi has welcomed Iran's recent announcements and emphasized the importance of resuming safeguards verification work after a 12-day conflict that damaged nuclear sites. Despite damage from military strikes, inspectors remain in Iran and are ready to verify nuclear materials, including enriched uranium. Grossi highlighted the need for cooperation to resolve Iran's nuclear dispute and reassured that there has been no radiological impact on the population or environment, although some localized contamination at sites like Fordow and Natanz has been identified. He also proposed a meeting with Iranian officials to facilitate cooperation.
Google is implementing more restrictions on its AI Overviews feature after it produced some inaccurate and unhelpful responses, such as advising people to put glue on pizza. The company has added safeguards to improve accuracy, including better detection of humor and satire, and limiting user-generated content in Overviews.
OpenAI's GPT-4 with vision, which combines text and image analysis, has been revealed to have flaws in a technical paper published by the company. While OpenAI has implemented safeguards to prevent misuse and mitigate biases, the model still struggles with making accurate inferences, hallucinates, and misses text or objects in images. It is not suitable for identifying dangerous substances or chemicals, and it misidentifies certain hate symbols. GPT-4V also exhibits discrimination against certain sexes and body types when safeguards are disabled. OpenAI acknowledges that the model is a work in progress and is working on expanding its capabilities in a safe manner.
Seven leading A.I. companies in the United States, including Amazon, Google, and Meta, have agreed to voluntary safeguards on the development of artificial intelligence. The companies will announce their commitment to these new standards at a meeting with President Biden. The safeguards aim to ensure safety, security, and trust in A.I. technology, as concerns grow over the potential spread of disinformation and the risks associated with self-aware computers. The companies have agreed to security testing, implementing watermarks to identify A.I.-generated content, publicly reporting system capabilities and limitations, deploying A.I. tools to tackle societal challenges, and conducting research on risks such as bias and invasion of privacy. The announcement comes as governments worldwide are working on legal and regulatory frameworks for A.I. development.
The Senate held a hearing on regulating AI, with OpenAI CEO Sam Altman warning about the dangers of the technology. Marc Rotenberg, executive director of the Center for AI and Digital Policy, emphasized the need for ethical guidelines and safeguards to limit the risks of AI, including algorithmic biases, disinformation, and societal inequalities. He also noted that AI researchers have expressed concerns about the technology's potential to cause human extinction and the need to address immediate concerns such as embedded bias and discrimination. The benefits of AI are widely acknowledged, but without proper safeguards and limits, the risks could outweigh the benefits.