AI competition heats up! Anthropic releases AI agents that can perform complex tasks using computers

Author: Zhao Yuhe

Source: Hard AI

Anthropic, an artificial intelligence startup supported by Amazon, announced a significant AI milestone on Tuesday: AI agents can now use computers to complete complex tasks just like humans.

Claude from Anthropic is one of the chatbots similar to OpenAI's ChatGPT and Google's Gemini. The release on Tuesday is an update to Claude 3.5 Sonnet, where the new "computer usage capability" feature in this latest AI model allows it to interpret information on computer screens, select buttons, input text, navigate websites, and perform tasks through any software and real-time internet browsing.

"It can use a computer in a way that is fundamentally similar to us," said Jared Kaplan, Chief Scientist at Anthropic, in an interview with CNBC, adding that it can perform "dozens or even hundreds of steps of tasks."

Anthropic stated that Amazon has been the first to test this tool, with early customers and testers including Asana, Canva, and Notion. According to Kaplan, the company has been developing this tool since the beginning of this year. Anthropic released the beta version of this feature for developers on Tuesday, and the team hopes to open it to consumers and enterprise customers in the coming months or early next year.

Future consumer applications from Anthropic include booking flights, scheduling appointments, filling out forms, conducting online research, and submitting expense reports. Kaplan stated:

"We hope Claude can truly help people handle various tasks, not just end after asking questions and getting contextual answers through chatbots."

In addition, Anthropic also released the next-generation model, Claude 3.5 Haiku. The company stated that Haiku now surpasses many state-of-the-art models in task completion, including the original Claude 3.5 Sonnet and GPT-4o, while maintaining the same cost.

The Rise of AI Agents

Since the global popularity of OpenAI's ChatGPT, the generative AI industry has rapidly shifted from text responses to generating AI photos, videos, and voices. Now, both startups and tech giants are fully engaged in developing AI agents.

Analysts believe that AI agents are not just about providing answers, but are designed for productivity, capable of completing multi-step, complex tasks on behalf of users. Although this term is not clearly defined throughout the tech industry, AI agents are seen as a step further than chatbots, typically designed for specific business functions and can be customized on large AI models Venture capital firm Lux Capital partner Grace Isford told the media in June that the interest of technology investors in startups building AI agents has "dramatically increased." These showcase companies have collectively raised hundreds of millions of dollars, and their valuations are also rising as the generative AI market expands.

Microsoft CEO Satya Nadella stated earlier this year in a financial earnings call that he hopes to provide an AI agent that can complete more tasks on behalf of users, although there is still "a lot of execution work to be done." Executives from Meta and Google have also indicated that they are pushing for AI agents to become more efficient.

Competing with OpenAI on multiple fronts

Since Anthropic released the first version of Claude in March 2023, the company has quickly become one of the hottest AI startups, directly competing with products like ChatGPT in the enterprise and consumer markets. Its supporters include Google, Salesforce, and Amazon. Since January this year, Anthropic has launched iOS and Android apps, a Team plan for enterprises, and expanded into the European market.

"We are moving towards a world where these models will be more like virtual authors than virtual assistants," said Anthropic's product manager Scott White in September.

Last month, Anthropic launched Claude Enterprise, its largest new product since the release of Claude, designed specifically for enterprises looking to integrate Anthropic AI. According to the company, early testers and customers of Claude Enterprise include GitLab, Midjourney, and Menlo Ventures.

Claude Enterprise allows customers to upload relevant documents with a much larger context window than before, equivalent to 100 30-minute sales conversations, 100,000 lines of code, or 15 complete financial reports. The plan also provides "activity summaries" for super users within the company to show new employees how others are using this technology.

In June, Anthropic also announced the launch of "Artifacts," which allows users to have Claude chatbots generate text documents or code and open the results in a dedicated window.

"Artifacts" or "workspaces" enable users to "view, edit, and build Claude's creations in real-time. This allows Anthropic's enterprise clients to create marketing calendars, import sales data, create dashboards or forecasts, write code for features, draft legal documents, summarize complex contracts, automate legal tasks, and more.

Shortly after Anthropic launched Teams in May, Mike Krieger, co-founder and former CTO of Instagram under Meta, joined Anthropic as Chief Product Officer. In addition, Jan Leike, former Head of Security at OpenAI, also joined the company in the same month

User Reactions

Some netizens expressed amazement at Anthropic's latest update, stating that they are witnessing a significant evolution in artificial intelligence technology, with autonomous agents becoming a reality.

Others believe that the naming of Anthropic is problematic. With such a major update, why not call it Sonnet 3.6?

However, some netizens are skeptical, believing that the new model does not have any advantages in creative work.

This article is from WeChat official account "Hard AI". For more cutting-edge AI information, please click here.