MICROSOFT LAUNCHES AI AGENTS

Plus Can AI Really Train AI? ▶️Meta's "Self-Taught Evaluator" trains and evaluates AIs autonomously, using the "chain of thought" method to boost accuracy, potentially outperforming human feedback.

Microsoft Announces New Generative AI Agents for Businesses

Microsoft has unveiled a series of semi-autonomous AI agents designed to improve business processes, from qualifying sales leads to managing customer interactions. Available through Microsoft 365 Copilot and Dynamics, these agents can work independently within set boundaries but require oversight to avoid unpredictable responses. Companies are exploring AI agents as a way to boost productivity, although they carry risks, including potential errors without human intervention. As generative AI evolves, Microsoft aims to integrate both agents and copilots, believing organizations need a balance between task automation and human assistance.

Nicolas Cage Warns Actors About AI Manipulating Performances

At the Newport Beach Film Festival, Nicolas Cage raised concerns about the use of AI and "employment-based digital replicas" (EBDR) in filmmaking, warning that AI could change actors’ performances without their consent. Cage urged actors to think carefully before signing contracts allowing studios to use this technology, citing his altered performance in The Flash as an example. He called AI "inhumane" and emphasized protecting what he termed MVMFMBMI: “my voice, my face, my body, my imagination—my performance.” This warning comes amid ongoing SAG-AFTRA strikes over AI use in video game production.

AI Bubble: Bigger and Riskier Than the Dot-Com Era?

Generative AI is experiencing explosive growth, but experts Jeffrey Funk and Gary Smith argue that it may end worse than the dot-com bubble. Unlike the internet boom of the late 1990s, AI adoption is fueled by free services that often lack meaningful economic impact. The costs of developing and sustaining AI—such as high-priced chips, electricity, and talent—currently outweigh productivity gains. With OpenAI anticipating billions in losses despite high valuations, the authors warn that AI may be an overhyped, unsustainable bubble, leading to a significant market crash.

Meta Unveils AI Model That Trains Other AIs Without Human Feedback

Meta introduced the "Self-Taught Evaluator," an AI model designed to train and evaluate other AIs autonomously. Traditionally, AI development relies on human feedback through Reinforcement Learning from Human Feedback (RLAIF), a process that can be slow and expensive. Meta’s new model leverages the "chain of thought" technique, breaking complex problems into smaller steps for better accuracy. According to researcher Jason Weston, as AI advances, it may outperform humans in reviewing and improving its own work.