OpenAI is working on an AI agent that's expected to launch in January. According to Bloomberg, the agent, codenamed "Operator" will be able to take over a person's computer and perform tasks for the the user like booking flights and writing code.
CEO Sam Altman hinted that this was coming in a Reddit AMA (Ask Me Anything) with other OpenAI executives, saying "the next giant breakthrough will be agents." Regarding future plans for more autonomous AI, CPO Kevin Weil added that the ability for ChatGPT to messages users first will be "a big theme in 2025." In September, users reported that ChatGPT was messaging them first. At the time, OpenAI said this wasn't supposed to happen, but it might be a sign of things to come.
Agents are shaping up to be the next frontier for the AI industry to tackle. Microsoft has AI agents for its Copilot model that can be customized by businesses to execute tasks on the user's behalf. Anthropic also released a feature for its Claude model that can take over a user's cursor and write code. And Google is rumored to be launching a similar tool, codenamed "Jarvis" for browsing the web, shopping, and booking flights for users. A leak caught by The Information showed this as a Chrome extension.
Meanwhile, reports from Bloomberg and The Information indicate that LLMs are hitting a developmental wall. Improvements of models are reportedly smaller and reaping diminishing returns due to fundamental limitations in generative AI architecture, despite scaling with more computing power, which continues to be expensive.
"Sky high valuation of companies like OpenAI and Microsoft are largely based on the notion that LLMs will, with continued scaling, become artificial general intelligence," said AI expert Gary Marcus, who predicted the wall in 2022. "There is no principled solution to hallucinations in systems that traffic only in the statistics of language without explicit representation of facts and explicit tools to reason over those facts."
Despite these reports, Altman said in the AMA that AGI (Artificial General Intelligence) is "achievable with current hardware."
Whatever the case, OpenAI has been focused on fleshing out more capabilities with existing models and shipping features that largely rely on variations of current LLMs.
