Google has formed a new AI team to focus on world simulation. Tim Brooks, a former OpenAI member, is leading this initiative at Google DeepMind. The goal is to create an AI model capable of simulating the entire physical world. Brooks describes the plans as ambitious, requiring massive generative models. Collaboration is ongoing with teams developing Google’s Gemini models, the video generator Veo, and the foundation model Genie. The team seeks employees skilled in solving novel problems and scaling models to their maximum potential.
Google researchers have developed a new variant of the Transformer architecture called “Titans.” This system equips language models with a type of long-term memory, allowing them to process information over extended sequences effectively. Important information is stored in separate networks within the Transformer, ensuring it is retained. A mechanism is also in place to discard irrelevant information, optimizing memory usage. In tests, Titans models, despite having fewer parameters, outperformed larger language models like GPT-4 and Llama3.1-70B in processing lengthy texts.
Apple is addressing issues with its AI summaries in iOS 18.3. Previously, Apple’s AI had inaccuracies when summarizing notifications, leading to calls for the removal of AI summaries. The latest beta version of iOS 18.3 no longer generates summaries for apps in the “News” and “Entertainment” categories. Apple plans to mark potentially false summaries in future updates to help users identify inaccuracies. A new warning in system settings now informs users that AI summaries may contain errors.
OpenAI is collaborating with a biotech startup. OpenAI CEO Sam Altman plans to present significant AI advancements, described as “superagents,” capable of solving complex problems with precision and creativity. OpenAI has also ventured into biosciences, working with Retro Biosciences to develop a model called GPT-4b micro, designed to optimize Yamanaka factors. These proteins can transform ordinary cells into stem cells, a promising approach for tissue rejuvenation and organ development. Initial results suggest the model’s proposed protein variants could significantly outperform human researchers.
The US Federal Trade Commission (FTC) is concerned about collaborations between major cloud providers and AI developers. The FTC is particularly focused on Microsoft’s $13 billion investment in OpenAI and the partnership between Google and Anthropic. The FTC fears these collaborations could lead to younger AI companies being fully acquired by tech giants. Amazon’s substantial investment in Anthropic is also under scrutiny. The FTC report highlights concerns about how partnerships between large tech companies could lead to market lock-ins, potentially stifling competition.
Google has signed its first content partnership for an AI chatbot with the Associated Press. This agreement aims to supply the chatbot Gemini with real-time news and current information. Similar agreements have been made by other tech companies like OpenAI and Meta with publishers and news agencies. Such exclusive deals could threaten media diversity if only selected partners are fairly compensated or prominently displayed in search results.
During the Maha Kumbh Mela pilgrimage in India, authorities are using facial recognition and AI technology to assist people who get separated in the crowd. Up to 450 million pilgrims are expected during the six-week event. Modern cameras, underwater drones, and autonomous buoys are used to monitor the massive gathering. Facial recognition and AI have already reunited over 250 individuals with their families on the first day.
In a global survey by the Boston Consulting Group, 73% of companies plan to invest in AI in the coming months. However, in Germany, only 65% of businesses intend to do so. More than half of German executives feel restricted by regulations when using AI, the highest rate globally. Additionally, 62% of German companies are concerned about data security, and 44% fear losing control. Despite these concerns, 95% of German executives do not expect a reduction in staff numbers due to AI. Instead, they plan to train employees in AI usage. However, only 30% of German companies have had more than a quarter of their workforce participate in such training, even though the AI Act will soon require it in the EU.