Top 3 Libraries for Crafting Powerful AI Agents (2024 Edition)

today's topic is about AI agents we'll talk about what's agents and what is the best framework you can use today for implementing AI agents there are so many out there we'll narrow it down to the top three for you so if you're just starting and don't want to waste time trying a whole bunch start with the ones that we recommend here today just to make it clear we're not affiliated with any of them that we're going to recommend today we just did the work and tried a lot of them and and sharing our experience with you today before we get started remember to subscribe to our Channel we'll bring you more content about what's the latest in AI how to get started how to transition into AI careers and so on awesome maybe let's start with what is AI agent there are so many different definitions of Agents if you take a look at different agent Frameworks and papers they have different definitions but more or less definitions are very similar so I use this metod GPT definition what they say agent in the context of large language models so an agent is an AI system or an entity which has some brain which is the llm itself it can observe it can think it can act and it has access to the memory so the combination of these five components form an agent this is another Library they have more or less the same definition an agent is an abstraction that encapsulates llm conversations State along with access to some memory and tools so again an agent is llm which can think observe has access to some tool and then it can act independently how is it different from uh rag system that we're familiar with basically rag is an agent if you think about it so the most like simplest right type of agent is CH GPT think about it J GPT is an agent why because it has brain which is the language model behind it it has some memory internal memory it can observe the observations are the signals that we give to that like the text the prompts these are observations the images things like that and then the the reasoning and thinking when we ask a question it tries to answer usually it uses internally some kind of thought process Chad GPT is an agent and rag again is is an application of using CH GPT with some external knowledge base right so rag is an agent if you think about it because in rag we are asking some questions about some proprietary data and then the llm tries to fetch some information from that data set and answer our question we are interacting with an agent and the agent has access to some external memory or knowledge base so that's why this whole abstraction the rag itself you can think about it as an agent so the AI agent agent may be more flexible than rag each one of them is a rag by itself it's going somewhere retrieving some data for the user to achieve some sort of task so if you stack them together then it can be a multiple rags and each one of them is an AI agent right yes so we call them agents because later on we would like to Define more complex tasks and give them to the agents to do that for us but the most simple agent is basically an llm itself and whatever you do you can call that an agent rag is one but typically we would like to use agents for more comp complex tasks for example when we are going to write a software let's say a game when we want to do some let's say research online for doing these types of tasks because we are not doing one simple let's say question and getting some response we are doing multiple steps agents are very good for these types of tasks when we are dealing with multiple things not necessarily one simple step and so that's why for those we call them agents and we don't usually call chat GPT in its simple form agent but in reality any AI system that has the features that I mentioned at the beginning you can call it an agent so L CH GPT then will be an agent yeah it's definitely more powerful than a rat right that's why it's called an agent it's using the analogy that it's almost like a humanik kind of system that go out and do something for you yes so you can Define so many different agents you can define an agent for doing only rag applications you can define an agent for example for doing research for writing a software so essentially whatever agent that you can think of as long as you essentially consider all of these components then you have an agent awesome let's go to the best agent library that we can use all right so I have found several different libraries agent Frameworks that you can use I experimented with some of them so one of the very good Frameworks for creating agents is this Library called meta GPT they published a very good paper several months ago in August then that paper in November they got accepted in the conference iclr which is really the top- notch conference so the paper is very well written it basically explains exactly how they designed this framework meta GPT in my opinion this meta GPT is the best agentic framework out there on their website they have a lot of documentation tutorials even different use cases they have agents for example for data analysis and visualization for doing machine learning for doing research different use cases just to give you an idea but obviously you can define a custom agent however you want as long as you follow the basic rules and objects and classes that they Define why why do you think this is the top one the number one the best one out there that's a very good question after experimented with several of these agent Frameworks you you made more sense to me that this their definition and the concepts that they Define in their library and use to create agents was the most relevant and the best in my opinion all of these libraries including metag GPT the concepts of agent is very similar however the way that they actually Implement that in their library is different and that's why meta GPT implementation of Agents is much better makes more sense to me that's why I choose metag GPT there is another popular library for creating agent which is called autogen here autogen this is also a very popular library for working with agents multi-agents and things like that it's backed by Microsoft they have also a lot of documentation and examples explaining different types of agents in different scenarios so the documentation is really good it's very thorough they explain different use cases they have Discord channel so autogen also is a very good Library I still prefer metag GPT over autogen because metag GPT is more flexible in terms of changing the behavior of the agent or if you want to create your own agent and let's say use your own implementation of Open Source llm models and different features and functions although you can use that with autogen I still like metag GPT implementation as I said it's much better in my opinion but autogen I have used it it's really good this is my again second favorite Library there are a bunch of other libraries out there one of them crew AI this is pretty recent this is also pretty easy to start if you want to create an agent or multiple agents using this library is very straightforward there are a lot of examples and documentation but the thing is the more abstraction they have added into to these libraries it's less flexible to have full control over the definition of agents and their behaviors and all that this library for example is not ideal it removes a lot of abstraction away from me this is probably for more non-technical users where they just want to create some agent and use it for a use case it is excellent but to essentially have 100% control I would rather use meta GP or autogen but this Library also has a lot of fan people are very interested in this Library crew AI there is another Library it's called open agents they also publish the paper this framework also is very nice I like it they have a very interesting documentation website with YouTube videos that explain how different use cases work and another another very interesting library is called chat Dev this is particularly designed for developing software if you want to create an agent for doing some research or something that is not necessarily software development related this may not be your best choice but if you want to do anything with software for example all of their examples that they use or know these agents are for creating games software applications things like that this is another very interesting Library we're at number five so yes yeah number six are we are we ranking this like by the order of how we are introducing them except for metag GPT and autogen the remaining ones they are not necessarily in order right from best to worst someone might like crew AI more than metag GPT also llama index yes they have this notion of agents and the good thing about llama index for example if you're using llama index for other applications like Rag and things like that then you can also use Agents from llama index because it's already integrated into the library but let's say if you want to use autogen then it's an external third party library that you have to integrate so it makes more sense depending on the use case to use the agents feature from the Library itself they definitely have this agent concept Lama index is one of my favorite libraries for working with llms and creating rag based applications Lang chain is another library that they have this notion of agent in there I tested one example just creating one simple agent using Lang chain in my opinion it's a very good Library however it has a lot of boiler plate and creating a lot of know adding a lot of code like things like that so I personally didn't like that idea but it's a very good library because if you're using Lang chain for other use cases rag question answering for other use cases if you want to create agents then you don't need to go somewhere else you can simply just use the agents the next library that has come into picture with very recently for again working with multi-agents and agent programming pretty recent haven't actually tested out this Library so I'm not sure how good or bad how easy uh it is to use this Library they have a lot of different guidelines and examples here and there are a few other libraries but these are pretty much the more common libraries out there that people are using right now yeah these are the really well-known ones right and we have narrowed down to at least the top three at least for us are metag metag GPT autogen and crew Ai and we're ranking from the perspective of flexibility a lot of it right the design of how agents is um structured and also how flexible it is for developers to be able to fine-tune your agents for your use cases So Meta GPT provides the most flexibility for our use case versus crew AI is very easy to use but may not have that level of flexibility so that's pretty much the criteria we are using to rank order these library but a lot of the other libraries that you introduced like llama index lenen those are very well know in libraries as well um depending on the use cases right users can can can check it out and see whichever one is suitable for your projects yes really yeah just to wrap it up if you don't want to have a lot of flexibility but you want easier to use kind of use case then Kuru AI probably is the best one autogen is not very hard to use and llama index if you want to have a lot more flexibility in terms of control designing and implementations then meta GPT and autogen are very good with these types of use cases yeah awesome and before we wrap up for today I one last question about agents for you do you think agents will take our jobs do you really think that's the future right because CAD Dev is the developer team basically it's for software de development I think our old engineer VP is going to be he already told me that he's not going to let his kid learn programming or computer science so it looks like yeah maybe this is why because he thinks that his kid is GNA run out of work yeah the Next Generation how are you gonna find work because you can hire a bunch of AI agents to develop softwares now and it can only get better right in 20 years think about it yes obviously it's very hard to predict what's going to happen in 20 years right AI has Tak the world with this storm and it is very unpredictable to see what happens it's just the beginning but eventually maybe most of the jobs will be done by agents or assistants right so we will probably focus on other stuff so I think more or less even now you can see a lot of tasks or jobs are being done by AI agents let's say chat GPT we don't call them let's necessarily agents AI assistants right but yeah I don't know I think that for a lot of jobs yes AI agents will take over those jobs maybe more jobs or new versions of jobs will be created so in general I am very optimistic but yes everything that is repetitive very tedious time consuming these types of tasks are going to be replaced sooner yeah you just reminded me I think we wrote about aogen at the end last year on substack and I attached link to our description and I was claiming an agent that will take your job so sounds scary but I think here's the thing if you imagine your employer is going to hire a bunch of AI agents to fulfill what they're trying to do today you can hire a bunch of agents to do what you want to do as well right which means you can start your own company you can realize your your idea more easily as well so it's encouraging for people who are very creative who wants to be entrepreneurs who wants to try something else who wants to pursue their own like hobbies and hobbies can become a business in the future and maybe the future companies are going to be really small because you really don't need a lot of you don't need a lot of people but it just encourage more Creations right that's what humans are best at right we should all start creating that's why check out our video and start creating try out agents libraries and then very soon you're going to be able to hire your own AI agent team for your own business ideas right absolutely I definitely agree just on a relevant note I this morning I saw I guess it was on Twitter some diagram shared by someone the number of AI startups compared to last year I guess it was at the beginning of 2023 let's say it was for example 2000 something like that but within just several months it had reached like 15,000 so it's like almost seven times more so yes uh a lot of jobs are going to be replaced but I think newer jobs newer form of jobs will be created I'm not really too worried about it right now yeah yeah that's a very empowering in my sense it's a little scary but I think it's very empowering as well the key thing is to how to face this wave of changes is anticipate that it's no longer information explosion it's going to be creation explosion but it's the going to be the next wave right you have to be ready to face a lot of new Innovations in a sense is a good saying I think I can be more productive in my daily life as well with all those AI tools definitely yeah I personally don't like to do a lot of repetitive tasks I would rather just automate them as much as possible and focus on other things that matter to me the someone will create that tool to automate your life for you I guess it's going to be me or otherwise I will be okay okay okay you have to automate we have to automate our lives or at least help our audience to automate absolutely yeah I agree I agree awesome all right talk with you live next time see you next time sure

Channel: TwoSetAI

Views: 1,570

Keywords: ai intelligent agents, #ai, artificial intelligence, #RAG, retrieval augmented generation, #innovation, #AIagents

Id: OskcinZOZ40

Length: 20min 53sec (1253 seconds)

Published: Thu Feb 29 2024