OpenAI has introduced updates to its Realtime API in London, enabling developers to incorporate voice features into their apps with new voices and prompt-generation functions, facilitating the creation of voice assistants. For consumers, OpenAI launched ChatGPT search, allowing internet searches via the chatbot. These advancements are foundational for developing AI agents capable of handling complex tasks, like booking flights. OpenAI aims to create agents itself and empower developers with its software to build their own. Voice integration will be crucial in enhancing the user experience of these agents, which can act as personal assistants by accessing emails, apps, and schedules. However, two significant challenges remain: reasoning capabilities and tool connectivity. OpenAI’s o1 model introduces a "reasoning" feature using reinforcement learning to improve task execution, though it still requires improvement in reliability, speed, and cost. Critics like Chirag Shah caution that current AI lacks true reasoning, often mimicking logical patterns from training data rather than understanding them genuinely.

OpenAI's future ambitions include expanding reasoning techniques to diverse fields beyond science and coding, such as law and economics. The second challenge involves AI models needing to connect efficiently with external tools for real-world tasks, going beyond reliance solely on training data. To address this, OpenAI's ChatGPT search allows models to browse the web for current information. Competitor Anthropic’s Claude chatbot offers a feature that interacts with computer interfaces, showcasing potential functionality for AI agents. Although OpenAI’s o1 can perform some tool interactions, further development is necessary. Godemont anticipates growth in AI adoption for tasks like customer support but acknowledges unpredictability in how users will employ OpenAI’s technology, expecting unforeseen applications. Additionally, an AI-developed version of Minecraft by Decart and Etched showcases real-time video generation through next-frame prediction, achieved by training the AI model on extensive Minecraft gameplay data, despite challenges like hallucinations. There are concerns about AI search reshaping the web, possibly threatening the digital economy by altering information dynamics. AI’s role in e-waste is highlighted, with potential contributions of up to 5 million tons by 2030 due to equipment used in generative AI models. Furthermore, AI's pitfalls were exposed in Poland when AI hosts inadvertently "interviewed" a deceased Nobel laureate, leading to backlash. Lastly, Meta CEO Mark Zuckerberg anticipates a surge in AI-generated or summarized content, predicting it will create an exciting new content category.