Source: Wall Street author: GE Jiaming

The second half of the big model battle may have begun, and AI Agent has become the focus of competition among the tech giants on the stage.

Artificial intelligence startup HyperWrite officially launched the AI Agent app on Aug. 3, hoping to become a "digital assistant" to human beings.

As an investor in HyperWrite, co-founder and CEO Aidan Gomez of Cohere, a generative AI start-up, tweeted: "We will begin to see real 'personal AI assistants' for the first time."

What is Agent? In the context of a large model, it can be understood as energy.Understand, plan and execute complex tasks independentlyDonovan's system.

AI Agents is not an upgraded version of ChatGPT, so it is more appropriate to think of it as a "digital assistant" for human beings. It not only tells you how to do it, but also helps you do it.

As can be seen in the introduction video released by HyperWrite，Users only need to give HyperWrite a purpose, and the rest of the digital assistant can help.

For example, throughChrome browser control program, HyperWrite can help users book air tickets, online celebrity restaurants, and even automatically help users order pizza takeout.

Wall Street has also tried HyperWrite to see if the assistant can help book a flight. The assistant explains each step to me, which is called thought process hint-- improving performance by helping LLM reason problems. Agents are suspended on websites that require personal information, such as login pages or credit card information.

In an interview with the media, HyperWrite co-founder and CEO Matt Shumer pointed out that ideally Agent can also log in, but this feature cannot be launched for the time being due to security and privacy issues.

In fact, the use of Agent in science fiction movies is not difficult, but in the exploration of artificial intelligence, the development of Agent has been going on for nearly half a century.

As early as the 1980s, computer scientists began to explore how to develop intelligent software that can interact like human beings. However, due to the limitations of data and computing power, AI Agents lacks the necessary realistic conditions.

Technology demonstration projects, represented by AutoGPT and BabyAGI, became popular briefly in April this year, but at that time the market thought it was still a long way from being actually applied to the business.

Just like Andrej Karpathy, co-founder of OpenAI and former artificial intelligence director of Tesla, Inc., shouted: "AI Agent represents a crazy future."

Now, the launch of Hyperwrite may mean that the second round of Agent outbreak is already brewing, and human beings may not choose this "crazy future".

The starting point of a new era of Agent--AI applications?

What is an Agent application? Weng Lilian, a Chinese scientist at OpenAI, gives the most intuitive formula:

Agent = large model + memory + active planning + tool use

Take AutoGPT as an example, a classic case is to enter a problem for a large model: identify an investment opportunity. Normally, a LLM cannot give a specific operation.

AutoGPT's idea is to first tell LLM about the problem, LLM gives several options, and then chooses a method. The AutoGPT itself can continue to execute based on the selected results.

After completing these tasks, AutoGPT continues to send the previous records to LLM and continues to ask for new solutions. This is a simple case of AI Agent.

Shumer pointed out thatHyperWrite is developing its own large language model "Agent-1"It is planned to use it to drive Agent in the coming month. At present, Agent is a combination of self-developed models and open source and non-open source large language models from investors such as Cohere.

According to Shumer, the problem with general-purpose LLM like Cohere or GPT-4 is that they are designed to accomplish everything from writing poetry to solving math problems, which makes their training slow and expensive. On the contrary
HyperWrite's homegrown model focuses exclusively on browsing the web, which means it is much simpler than other LLM, so it is faster and cheaper.

In March and April, a number of AI Agents broke out, such as Camel, AutoGPT, BabyAGI, West World town and so on, which seemed to make people see the possibility of application landing.

Less than two months after Significant Gravitas opened up AutoGPT in March, AutoGPT has obtained 130000 star on GitHub, making it the fastest-growing open source project in history.

Sam Altman privately told some developers in May that OpenAI wanted to turn ChatGPT into a personal work assistant, and a person familiar with the matter pointed outOpenAI has been focusing on how to use chat bots to create autonomous AI Agents, and related functions are likely to be deployed in ChatGPT helpers.

Coincidentally, Meta sees an opportunity for AI Agents.

Back in April, Mr Oza told investors that Meta saw "an opportunity to introduce AI Agents to billions of people in a useful and meaningful way", but he did not specify the application.

At a plenary meeting with employees in June, Xiaoza announced a series of technologies at different stages of development, one of which is to bring AI Agents with different personalities and abilities to provide help or entertainment, initially mainly for Messenger and WhatsApp.

The starting gun of the AI Agents track has been fired, who will be the leader of the track.

Edit / lambor