I Tested a Next-Gen AI Assistant. It Will Blow You Away – WIRED

Create a 3:2 aspect ratio illustration in a cheerful, cartoonish and vibrant style, showing a non-descriptive male software engineer adjusting a large, complex graphics processing unit. In the background, show data being processed and transformed into artificial intelligence sub-systems, representing various countries' tech advancements. The overall image should have a positive and light-hearted tone to represent the growing demand for such technology.

The rise of AI-powered chatbots like ChatGPT and Google Bard is making virtual assistants like Siri, Alexa, and Google Assistant seem less impressive. However, when the generative AI technology behind these chatbots is integrated into legacy assistant bots, they will become much more interesting. An experimental AI voice helper called vimGPT, built on OpenAI’s GPT-4V language model, showed promise in tasks like subscribing to WIRED magazine online. While legacy assistants handle simple tasks like setting reminders, buying something online requires more complex skills like understanding requests, navigating the web, and interacting with forms. VimGPT, although lacking credit card information, excelled at searching for cat videos and finding cheap flights. VimGPT is an open-source program built by a lone developer, but companies like Apple and Google are likely conducting similar experiments to upgrade their own assistants. Experts believe that the next step for AI assistants is to become agents that can perform useful tasks and solve problems for users. Simulated environments like VisualWebArena are being used to test and improve the skills of AI agents, but there are still challenges and bloopers. While AI agents can achieve complex objectives about 16% of the time, humans succeed 88% of the time. Despite the challenges, researchers are optimistic about the potential of AI agents to make digital life easier.

Full article

Leave a Reply