AI Models and Tools: OpenAI Launches AI Agent
OpenAI unveiled Operator on Thursday (Jan. 23), an artificial intelligence (AI) agent that can use a computer like a human, even using a cursor and keyboard. It is able to interpret screenshots and interact with graphic user interfaces (such as buttons, menus and text fields) on the computer screen, OpenAI said in a blog post announcing the AI agent. Through a browser, users can ask Operator to do things like order groceries, book reservations and buy event tickets, among other tasks. OpenAI noted that users will still have to approve sensitive transactions, such as making purchases, doing financial transactions and sending emails.
Advancements in AI Technology
“This represents an important step towards a future where ChatGPT is not only capable of answering questions, but can take actions on a user’s behalf,” according to OpenAI’s system card on Operator. However, OpenAI also warned of risks that include malicious instructions from third-party websites could mislead the model or let it do harmful or banned tasks, as well as ChatGPT making mistakes that might be hard to reverse.
OpenAI used supervised learning on specialized data and reinforcement learning to enable Operator. Supervised learning was used to teach Operator how to read the computer screen and click on items accurately. Reinforcement learning teaches the model reasoning, error correction, and the ability to adapt to unexpected events. 
A preview of Operator is available initially to U.S. ChatGPT Pro users.
Competitors in the Field
OpenAI’s move follows in the footsteps of its competitors. Last October, Anthropic announced a similar capability for Claude 3.5 Sonnet, an AI model that is part of its flagship LLM family. The OpenAI competitor called the capability “computer use.” However, Anthropic warned that the experimental features are “at times cumbersome and error-prone.”
In December, Google announced its own web-browsing AI agent, called “Project Mariner.” Built with Gemini 2.0, Google’s latest version of its flagship multimodal models, Project Mariner can “understand and reason across information in your browser screen, including pixels and web elements like text, code, images, and forms.”
Partnership Changes
Microsoft will no longer be OpenAI’s exclusive cloud provider for its AI models, the software giant disclosed in a Tuesday (Jan. 21) blog post. Instead, Microsoft will have the right of first refusal to host OpenAI’s AI workloads in Azure. This is a change from their 2019 agreement, when Microsoft became OpenAI’s exclusive cloud provider after investing $1 billion in OpenAI.
The investment was also three years before ChatGPT and generative AI took the world by storm. Since then, Microsoft reportedly has invested a total of nearly $14 billion in OpenAI. However, Microsoft will retain the exclusive rights to OpenAI’s application programming interface (API).
Future AI Projects
The change in terms come OpenAI on Tuesday named Microsoft as a technology partner in its new AI infrastructure project called Stargate, which aims to spend from $100 billion to $500 billion to build physical and virtual AI infrastructure. This includes the building of 500,000-square-foot data centers for AI processing. Ten data centers are under construction in Texas and will expand to 20. The first one is being built in Abilene, Texas.
Fresh off its success with NotebookLM, which turns documents into audio podcasts featuring two AI-generated hosts discussing the content, Google has opened the waitlist for its spinoff, Daily Listen. Daily Listen brings the two AI podcast hosts back. This time, they will give you a daily update on things that matter to you, based on your interests. You’ll also get links to stories from around the web. For now, Daily Listens is only available in the Google app on your mobile device, in the U.S.
Microsoft researchers recently introduced an AI model that can create new, inorganic materials with specific properties. It can do so much faster than the traditional way, where scientists would spend years in research. Called MatterGen, the AI model could lead to breakthroughs in things like batteries, magnets, semiconductors, and other technologies by creating materials that are more efficient, stronger, or cheaper, for example. 
MatterGen could also help tackle sustainability challenges by creating eco-friendly materials with less environmental impact. It can design materials that don’t rely on expensive or rare elements to reduce environmental harm. Scientists can use MatterGen to customize materials for nearly any purpose — whether to build stronger buildings, make better electronics, or improve medical devices.
