Overview of the current AI trend and market
The rapid development of artificial intelligence (AI) technology has been seamlessly integrated into our daily lives, revolutionizing the way we interact with the world. DIN is at the forefront of this revolution, aiming to build an intelligent network powered by data and AI-Agent. In the following section, we will skip the stage of general AI and go directly to the discussion of large language models (LLM) and AI-Agent. The reason is very simple. Whether in academia or industry, the development of AI will eventually point in one direction - AI-Agent. Humanity has entered a new era - the era of AI-Agent. In this era, AI has transcended its previous role as just a tool or service provider; going further, it has become an intelligent agent with the ability to proactively learn, adapt, and autonomously undertake complex tasks.
History of AI-Agent:
In the 1950s, Alan Turing extended the concept of "highly intelligent organisms" to artificial entities and proposed the famous Turing test. This test is a cornerstone of AI and aims to explore whether machines can exhibit intelligent behavior comparable to humans.
These AI entities are often called "agents" and form the basic building blocks of AI systems. So far, the Agent mentioned in the field of AI usually refers to an artificial entity that can use sensors to sense its surrounding environment, make decisions, and then use actuators to take response actions.
With the development of AI, the term "Agent" has found its place in AI research to describe entities that display intelligent behavior and have qualities such as autonomy, reactivity, initiative, and sociability. Since then, the exploration and technological progress of Agent has become the focus of the field of AI.
The late 1950s and 1960s were the period of the creation of AI, and the programming languages, books, and movies that emerged are still influencing more people today.
After experiencing the first AI winter, there was an AI boom in the 1980s. Various types of research during this period have made breakthroughs, investment from the government and other institutions has also begun to increase, and researchers have gradually increased their exploration of AI Agents.
But this craze only lasted for 7 years, and in 1987 we ushered in the second AI winter.
This cold wave lasted for many years. Although most institutions lacked financial support during this period, AI still developed resolutely along the existing technical lines.
Among them, AI Agent was defined by Wooldridge and Jennings in 1995 as a computer system: that is located in a certain environment and can act autonomously in this environment to achieve its design goals. They also proposed that an AI-Agent should have four basic attributes such as autonomy, reactivity, social ability, and initiative.
After AI Agent was officially accepted by economics, it was further defined as a system that can perceive its environment and take actions to maximize the chance of success. According to this definition, a simple program that can solve a specific problem is also an "AI-Agent", so later robots that can compete with humans in various chess games are also considered a type of AI-Agent.
During the period when AI Agents were given "four basic attributes", from 1993 to 2011, many impressive Agent-type projects based on the AI technology of the time appeared.
The appearance time and introduction of these projects are as follows:
1997: Deep Blue (developed by IBM) defeats world chess champion Garry Kasparov in a highly publicized match, becoming the first program to defeat a human chess champion.
1997: Speech recognition software (developed by Dragon Systems) is released for Windows.
2000: Professor Cynthia Breazeal developed the first robot that can simulate human emotions with its face, complete with eyes, eyebrows, ears, and mouth, called Kismet.
2003: NASA lands two rovers (Spirit and Opportunity) on Mars, where they navigate the surface without human intervention.
2006: Companies like Twitter, Facebook, and Netflix begin leveraging AI as part of their advertising and user experience (UX) algorithms.
2010: Microsoft launched the Xbox 360 Kinect, the first gaming hardware designed to track body movement and translate it into gameplay directions.
2011: An NLP computer called Watson (created by IBM) programmed to answer questions beat two former champions of the televised quiz show Jeopardy.
2011: Apple releases Siri, the first popular virtual assistant.
From these classifications and basic definitions, many AI tools and early intelligence programs can be classified as a type of Agent. Including the early IBM Deep Blue used for chess games and AlphaGO, which will appear later, they are all AI Agents based on the latest AI technology at the time.
With the development of AI technology and the continuous enrichment of data, AI-Agent has also evolved to a stage based on large language models(LLM).
2012: ImageNet Computer Vision Challenge, the deep learning model of the AlexNet convolutional neural network won first place, and deep learning has truly shown its prowess in the field of AI.
In 2016, AlphaGO (Google's AI Agent specializing in the game of Go) defeated the European champion (Fan Hui) and the world champion (Lee Sedol) and was quickly defeated by its brother (AlphaGo Zero).
In 2017, Google proposed "Transformer".
In 2018, Google released BERT based on the Transformer model, kicking off the big language model.
In 2019, Google AlphaStar reached Grandmaster in the video game StarCraft 2, outperforming all but 0.2% of human players.
In 2019, OpenAI released the natural language processing model of GPT-2 and released GPT-3, DALL·E 2, and GPT-3.5 in 2020 and 2022 respectively. The popularity of ChatGPT has paved the way for the development of AI Agent in the era of large language models. and applications provide new opportunities.
Starting from January 2023, global manufacturers have released multiple LLMs, including LLaMA, BLOOM, StableLM, ChatGLM, and many other open-source LLMs.
At the same time, thousands of LLMs launched by global technology manufacturers provide a broader foundation for the diversified applications of AI Agents in various fields.
On March 14, 2023, OpenAI released GPT-4. At the end of March, AutoGPT was born and quickly became popular around the world.
Auto GPT is a free open-source project launched by OpenAI on Github, which combines GPT-4 and GPT-3.5 technologies to create complete projects through APIs.
Different from ChatGPT, users do not need to constantly ask questions to the AI to obtain corresponding answers. In AutoGPT, they only need to provide it with an AI name, description, and five goals, and AutoGPT can complete the project by itself. It can read and write files, browse the web, review the results of its own prompts, and combine them with said prompt history.
AutoGPT is also an experimental project of OpenAI to demonstrate the power of the GPT-4 language model. Since then, more people have gradually become aware of AI Agent while understanding and experiencing AutoGPT.
Since then, AI Agents based on LLM have sprung up like mushrooms after rain, and many projects such as Generative Agent, GPT-Engineer, BabyAGI, and MetaGPT have emerged. The outbreak of these projects has brought the development and application of LLM into a new stage and has also brought the development of LLM to a new stage. Entrepreneurship and implementation lead to AI Agent.
In May, after OpenAI raised a new round of US$300 million in financing, founder Sam Altman revealed that he was paying more attention to how to use chatbots to create autonomous AI Agents and would deploy relevant functions to the ChatGPT assistant.
In June, Zuckerberg announced a series of technologies in various stages of development at an all-staff meeting. One of them is the release of AI Agents with different personalities and abilities that can provide users with assistance or entertainment functions.
At the end of June, Lilian Weng, head of the OpenAI Safety team, published an article called "LLM Powered Autonomous Agents", detailing the AI Agent based on LLM, and believed that this would turn LLM into one of the ways to solve general problems. .
At this point, people finally have a comprehensive understanding of AI-Agent, and the mystery of AI Agent has finally been unveiled. The exploration of AI-Agent in the field of AI has never stopped. After each AI technology achieves a new breakthrough, organizations will incorporate its exploration and application into new topics. After deep learning and neural network technologies represented by AlphaGo came to prominence, agents based on deep learning and neural networks appeared and were used in many fields such as games and medical care.
In recent years, large language models have achieved breakthroughs. After Google released Bert and OpenAI released GPT-2, many organizations began to cooperate with them to build agents based on LLM.
While we are still talking about AI-Agent, many AI Agent frameworks and products have appeared around the world. For example, Voiceflow, which just completed US$15 million in financing at the end of August, is now one of the most popular Al Agent building platforms among developers, with more than 130,000 teams efficiently collaborating here to build their own Al Agents.
Judging from this type of AI Agent construction platform, many organizations are currently building or have built their own AI Agents, and each organization can target multiple Agents for different business scenarios.
AI-Agent market status:
Demand for Agents will grow as they save humans the scarce resources of time and money; but the more obvious truth is that the infrastructure layer that enables the deployment of Agents is still nascent, with most Agent frameworks, such as BabyAGI, being experimental open-source projects. As the demand for Agents grows, so will the need to formalize these foundations. Each sub-segment will be productized (made into an app or API) over time. Humans will eventually be able to create or hire Agents as a product.
According to Annie Liao's opinion, the AI Agent can be categorized into three layers, which are
AgentOps: There are seven core components to this layer, according to Weng, Lilian's theory. These combine to form “frameworks”, which will be templatized and eventually listed on an Agent Ops-like marketplace for distribution.
Applications: Agents will “become products”. As the landscape matures, many applications will be productized and monetized through closed or open-source models.
Services: provide end users value-added capabilities like creating, trading, and maintaining the agents. Specifically, there is a list of Agent projects in the Crypto domain. Compared to previous projects, although these are in earlier stages and are primarily narrative-driven, we can delve into them to explore how AI and crypto are integrated.
Future
The Agent will become the main form of extensive model application in various industries and fields. In the future, the development and application of LLM will be presented in the form of tools or assistants around Agent. As Agents appear in the form of standardized products, it will become easier for organizations to introduce and apply AI Agents.
Relevant enterprises and organizations can build domain-oriented Agents based on the introduced large language model or vertical domain model to help customers efficiently release the capabilities of LLM. You can also build internal or customer-oriented AI Agent platforms and communities to facilitate your own and customer operations to build the required Agents at any time.
More AI Agent construction platforms will promote the emergence of a large number of Agents, and it will also be easier for individuals to build and apply Agents. In the future, as long as everyone is willing, they can create their own personalized Agent through various Agent platforms at any time, enhance communication and collaboration, and expand knowledge and skills through more personalized functions and services.
It is even possible to build multiple different Agents in different business scenarios and let these Agents work together. Multi-Agent system collaboration can output more accurate results and complete more complex tasks.
AI Agent ignores industries and business scenarios and can build corresponding Agents wherever LLM can be applied. It can be applied to various industries, such as education, medical care, finance, crypto, manufacturing, entertainment, etc., to help improve efficiency, reduce costs, and create value.
In the future, AI Agents may be more intelligent, adaptive, and diversified, able to handle more complex problems and scenarios and form closer cooperation and symbiosis with humans.
With the widespread application of AI Agents, human-computer interaction in the era of large language models will also be upgraded to an automated cooperation system between humans and AI Agents. This new type of human-machine cooperation can be called human-machine intelligence. It will promote the further upgrading of the production structure of human society and then affect all aspects of society.
At the same time, an intelligent network with the ability to communicate and perform tasks autonomously/automatically will become the next stage of the Internet, and AI Agent will be an intelligent tool for humans to interact with and perform tasks.
In the future trend, AI agents will most likely appear in various scenarios such as human work, study, life, and entertainment. Everyone will be equipped with an intelligent assistant based on the AI agent system, such as "Iron Man", "Interstellar" and "Star Wars" The scenes of human-machine collaboration in movies will truly become a reality.
This will be a market of many magnitudes.
最后更新于