Commercialize LangChain

Key Takeaways

hello community let's talk about something a little bit controversial Langton and you know it's not often that there's an exact moment that I know when I decide to make a new YouTube video but this is here the moment it was here a tweet by Max wolf and he said I have a full-on problem with length chain I spend a month working with link chain and Taft yes yes yes bullet points you can read it yourself and it's a debugging a link chain error is nearly impossible and if you need anything outside the workflow in the documentation it is extremely difficult to hack even with custom agents the documentation is missing a lot of relevant detail and the extreme popularity of link chain is warping the entire EI ecosystem around the workflows to the point of harming it so whenever you harm an AI ecosystem you know the recent releases by hugging face and open Mei recontextualize themselves around length chain it's just magic AI to the point of hurting development and hurting the code clarity and he says Hey part of the reason I'm hesitant to release this blog post is because I don't want to be that a-hole who criticizes open source software that's operating in good fate and I was amazed I said interesting what is going on with link chain and it took just exactly two days and I got this here in my Twitter feed from my user said oh great another a-hole who criticizes open source software that's operating in good faith and then the inside link chain indeed so you see there's a non-performance balance chain even the heavily critics notice there is something with flank chain so what is it and what is this social pressure that we are not allowed to talk about some open source project suddenly we cannot criticize this we cannot be helpful in further developing the system what is this social pressure so hey why we have a Woman by the way to indicate social pressure hey GPD find me another picture GPD this is again a woman another picture hey come on you want to tell me that we only have women here when when we depict social pressure and man okay man I mean a real a real man does not feel social pressure he exchange social pressure another one okay okay okay I can learn no problem so um Dear Community today I am a woman so hey to all my haters out there in the internet today I'm smart I'm fragile I'm weak so come and get me now you have here the picture of my inner self come on now it's too late sorry now back to the topic so here we are but even here article into a data science just two days ago when I record this video you know there's this other documentation of Langston is really good however there are not a lot of examples of some specific things my goodness why you don't try to say hey the documentation is missing hey do something okay now I read hack and use and I think it's a very valuable source of information and you might be a little bit offended by the word hacker but it gives you really some deep insights into people who are so much more involved in coding that I am so I Rely quite a lot on their inside so four days ago the reason why link chain is pointless is that it's trying to solve problems on top of technical Foundation that just cannot support it and if you want to read it there are 176 comments in four day so go to Hacker News not to my comment here in this video go to Hacker News and there you can provide the community with your excellent and exquisite knowledge but you know this is not a phenomenon of our current times because as you can see here I can use about 70 days ago they only discussed this and it says it was instantly deflated here when he looked at exactly the way how it works the common is I got that itself got a lot of traction on non-coders and I think this is the important thing of people who are not familiar how to code python thinking it's doing some magic but if you read The Source it boils down to some riddle prompts so you see there is a hype with Lang chain four non-codas opening up if an encoders the beautiful world of coding which is great in a way that we have some tools that help people learn how to code get familiar with more complicated topics but at the other side there's also an imminent danger if people just trust what is built for them and are not critical are not evaluating the solutions and just take what they get for free so beautiful yes something other there here the Langston introduced too many constraints in comparison to what it provides the built-in agents are close to useless for anything but for toy projects so you see there is a lot of criticism go here I have here the link for you and start or continue in the discussion with your sink yeah let's have a video I asked gpd4 to analyze what are the current drawbacks of flying chain as posted by people on the internet so first is outdated examples some users complainted example provided in the documentation use outdated version of Lang chain which can lead to confusion and difficulties when trying to learn how to use it the second point is the most important I think while the high level of abstraction can simplify application development it can also make the framework more difficult to understand it especially for beginners and I will talk about this in a second about documentation now for an open source Community project this is this is to be expected the documentation has been criticized for being poorly organized and lacking Clarity this makes it more difficult for users to understand how to use length chain and how to contribute to the open source project and the community itself some users have expressed dissatisfaction with the lack of community gathered on searchable forums and I said explain what you mean with the high level of of abstraction in theory and then provide two examples for a deeper Insight because I want to understand what is this abstraction and gbd4 comes back and says in programming abstraction refers to the process of simplifying complex system by breaking them down into more manageable parts so high level abstraction means that the details of the underlying implementation the code implementations are hidden and users interact with the system through a simplified interface and I think this is maybe here the the main root cause why this is happening so this allows users to focus on the problem at hand if they want to develop an app for example rather than getting bogged down in the details how the system works but this is exactly the problem if the system does not work for your specific case anymore if you have to debug or if you have to find out this very tiny little mistake or error that is happening somewhere in the system so in the context of Lang chain high level abstraction means that the framework provides a simplified interface for working with llms users do not need to understand all the details of how these models work they can simply use the tools provided by link chain to develop application and this is great as long as those tools provided by link chain work immediately out of the box then the world is beautiful but what happened if there is a problem and yeah the two examples data awareness one of the key features of line chain of course is its ability to connect to their llm models to other sources external sources of data this is indeed a complex process that involves fetching data from The Source like a database or something else processing the data cleaning the data and integrating the data with the language model specifications however lime chain abstracts away these details allowing users to Simply specify the data source and let the framework handle the rest and this works great until it doesn't work anymore chains line chain allows users to chain together different components to create more advanced use cases this involves coordinating multiple components each with its own complexity so you can add complexity on complexity and as long as it works and as long as this specific code is exactly targeted on your use case this is a fantastic development however Lang chain provides a high level interface for creating and managing these chains hiding the complexities of the underlying components and I think this is the main root cause hiding the complexities is great for beginners until you start to debug the system for example a user might create a chain that involves fetching data from a source processing it with a language model and then take some actions based on the model's output despite the complexity of this process the users interact with it with simple high level interface provided by link chain and now comes the part that amazes me GPS this example illustrates how LinkedIn uses high level abstraction to simplify the process of developing applications with llms or four LMS but by hiding the complexities of the underlying system if users do not understand what is happening under the hood of Lang chain link chain allows users to focus on what they want to achieve rather than how to achieve it so everything is fine if the car starts in the morning but what you do when it's a rainy morning you enter the car you want to start a car and the car does not work so do you understand how to achieve your goal if you have to start the debugging length chain can you expect this from an open source model talk about this so today I found here in my Twitter feed another tweet and I think it's very interesting so let me have here a look so it's about Lang chain and the problems I see with it currently and how I think it could improve so we have here user says hey I'm working with this I have some ideas how to make it even better so there's a heavy use multiple layers of abstraction make it really hard to tell what is going on under the covers to debug or to modify the code slightly this is the biggest thing I've seen people struggle with struggling with myself it is just not clear what exactly is going on a lot of the time and this cannot be it if you have here coding system that you do not know what's going on for example though yes yes yes you can read it yourself great second Point most of the connectors to external servers are Half Baked but this is this is evident listen those are people who who spend their free time helping other people to code so those professional coders who are so grateful and they donate their time here to build these systems of course the the code is not absolutely ready it's not absolutely 100 validated as not an absolute perfect documentation those are people who want to give you the tools so that you are able to understand to learn and you go on from those steps but I think the misinterpretation is that this is a commercial tool although it is open source it is 100 operational and people get frustrated a lot of yes yes yes but people don't necessarily realize that in the end up frustrated trying to use them yeah again hard to debug this kind of goes back to item one but I want to call it out specifically when something goes wrong it is really really hard to find out where the issue is mostly because of the layers of abstraction Yes again yes yes yes you know this the docs currently have some good example on how to use out of the box it would be nice to see some dogs that discuss more advanced some app focused some real world use cases so very clear message and he says hey you guys have done more than anyone to move this ecosystem forward and have done so at bracknack break neck speed this necessitates a lot of experimentation in shipping as fast as possible but I think that we now need is a more robust a simpler version that only provides a concise bare minimum set of tools that a Dev needs to ship llm apps I personally would love to see a version of length chain that is essentially just a collection of function and structures rather than layers of abstraction and connectors to external tools so you see there is now let's look at this at the point of an innovation system Let's do an abstract View if you are a lightning Feast fast with code implementation code implementation and the next code implementation and you do not in parallel optimize your documentation for users how to use it if you do not explain with tutorials if you do not explain with a lot of examples to users different real world use cases okay there comes a point when all your lightning speed coding and coding and coding is not effective anymore because people can't actually apply the code for their problem so I think a highly interesting tweet and yeah Mr Mr Lang chain himself personally says hey this is awesome and much appreciated feedback and you have also hey thanks for the feedback we just added some use case specific documentation as you suggest here from Lance Martin and here by Lance Martin here this picture I think it tells you a lot off about langchin it is rather easy look you have data connectors everything from public here too to private closed section unstructured data structured data everything you have and then you have to Simply a transformation you have a vector storage with 40 Integrations 50 integration 100 integration and then you get some document documents out some some strings some numerical data and this is the input to a gpd4 to a GPD 3.5 to any llm and this llm gives you an answer so you see all that Lang chain does here in this ecosystem connect you to different sources of information thus simply a cosine similarity operation a similarity operation in a vector space using the vector storage retrieve some docs and then you have llms like two before that does the real work where the real intelligence is I will have an upcoming video where I tell you that with llms having almost infinite token length and I will show you an example of Microsoft long net where they say they have a 1 billion token length for their llm you do not need Vector storage anymore you do not need this anymore if you have one billion token link like Microsoft claims in its latest scientific preprint you can get rid of all the vector storage and all the vector databases and have a complete new chapter of this topic but this would be Beyond today's video they are I have to say amazing people amazing Coda Jose here a tweet in reply to the first tweet I showed you hey there I'm with you on all points so I built light chain so he built your own system that knowing now here like Jane and its shortcomings he optimizes this to a new system and I have to say this is the real power of Open Source if you see that Samsung is not working if you are missing something and you understand what the current system is you build a better system this is amazing congratulations I'm just trying to get familiar with light chain there is a GitHub I can recommend it I just had a look at it today so I'm really at the very beginning light chain is a light alternative to length chain for building llm applications and instead of having a massive amount of features and classes light rain focuses on having a single small core easy to learn easy to adapt well documented fully typed and truly composable oh you do it pip install light chain let me show you in a video so let's see here short Jupiter notebook pip install light chain this is all there is to it and then you just have here open AI Jane yes slide tune installs beautiful and then you have just your open AI chat chain and you can build here your emoji chain for example and the funny thing is you can then Define a translator chain and you get it you connect the two chains together the Emoji chain and then the translator chain and so you have it what is really nice about this is what I like to show you is the documentation light chain is a lighter alternative to Lang chain for building LMS yes yes yes getting started as I just showed you you have to have your open EI key of course and then they have a lot of examples and whatever and I explained to you the basics why we need here streams async working with streams composing here to change the different chains with llms they explain you even then our function calling functionality of openai API you can add some memory to it in a very simple simple text completion memory so you have here all options and a really interesting documentation why not give this a try but there is also another alternative you remember with the tweet at the very first page of this YouTube video now this order also Max wolf set out to do his own system and he called it simple AI chat and he says yeah I built this python package for easy interfacing with chat apps like try trippity gpd4 with robust features and minimal code complexities and he goes on and explain this so I'll leave you here the GitHub great he also explains why he did this after working here for a long time with length chain he saw its time for a more simple implementation that is easier to maintain and resulted in a better generation here for prompts for the llm so also reading Hacker News if you don't know them I think they are really an interesting source of information so another system that substitutes augments Lang chain so if you're a beginner and using length chain is the one and only thing in the universe hey time to learn so let's come up to a summary I think what is really interesting is there's a great open source effort by Lang Chain by alternative system to link chain and I think we as users of python libraries and on we have maybe two check our expectation mindset and you'll say Hey what if what if link chain is developed so fast which is great that may be the documentation is suffering and I noticed for myself if I code for myself here when open source and not a client work I am not perfect in my personal documentation if I document my code what if Langston is developed so fast that's the validation of the code sequences is not perfect you just say is it working yes no continue next topic what if link chain is coded by professionals who are great but given their limited time and they have maybe a job and they have maybe a family that they have to care for what if those professionals did not cover every possible use case did you as a user might encounter so you have to prepare yourself it is not a complete perfect system what if LinkedIn is there to show you one possible solution but you have to code your own use case you have to find your own intelligent solution giving your personal ecosystem giving you company ecosystem imagine you have to optimize this interface functionality with your own python coding but you have a path you have a guiding system which is great in itself I have a video on gbd4 code interpreter and I think it is interesting to compare those two systems and you can say Hey what if length chain could learn from gpd4 code interpreter by open EI that goes step by step with clear precise explanations generates the python code for you shows you the code explains the code to you in detail and if the code is not working provides an alternative code solution to you if you think this is impossible have a look at my video where I show you code interpreter because this this way this particular way of finding a solution enables you to learn and to advance your knowledge and I can tell you I had to look at both systems and code interpreter is a really nice implication implementation to understand how to form a latest solution given a complex system understanding how to code it efficiently and of course we all learn continuously so there you have it now coming back now to my title how can we improve an open source project like Lang chain are we not allowed to critique an open source project because we are simply afraid if we critique it that people will get demotivated and stop providing us with free code should we really escalate the system to a point where we cannot use this open source code anymore because we do not understand it we have a level of abstraction we are not able to cope with and there is a missing documentation now looking at it from the point of an innovation Theory it is interesting because it is suffering here the open source ecosystem exactly from the points that a commercial entity would focus on because a commercial entity would understand that those pointer is essential for Revenue generation you have a professional support and a commercial entity would establish a dedicated support team to assist users with the issue from simple queries to providing some in-depth technical assistance now you cannot ask this from an open source community maybe you can have a platform but you cannot have that you know you can call a number and you get a solution for your company problem now for the maintenance of code a commercial company understands that the maintenance of CODIS is essential for each and every user that pays for that so it could employ a team of developers to maintain and improve their code base fixing bugs optimized performance implementing new features but at the same point and this is point number three have the perfect documentation training materials for user to learn how to use the software and get them at most of it for their job you could promote the software to a wider audience increase its visibility attract more user to your libraries you could reach out to potential Partners to sponsors secure additional resources for the project significant point is quality assurance if you read Reddit or go to Hacker News the quality that your code has to provide is essential people just use it as it was noticed just for toy project and the code breaks apart if you use it for a serious commercial undertaking for building a specific app based on another lamp if users get frustrated because it's not working quality assurance is for any commercial entity a point where they focus their resources on but you cannot ask an open source community to focus on quality assurance in the same way a commercial entity would or can you what do you think about this and finally Financial stability you have to have the financial resources to get the personal resources to get to professionals on board that take care about all the points we just had a look at in this sheet you need to have really the resources that needs to continue develop this improve the software make it accessible for everybody and not just jump ahead with code development and users cannot apply the code because they do not understand what's going on they cannot debug it they don't understand what's going on this level of abstraction actually hinders the further development of this open source community so there's a lot to learn from this and I think we should be allowed to talk about possible ways how to further optimize here a open source in our community if commercial involvement shows us how things could be done I don't say that we should do it in that way but we should learn from this we should learn what is missing it would be highly interesting here from Innovation Theory standpoint how would an open source Community evolve to take care about these topics and not become a conversion a commercial entity so you see there are a lot of open questions that we as a user and those participating here in the open source community of Lang chain will be faced in the future but we can address shortcomings we can congratulate them if they do some great work and if you want to try to keep the project open source and of value for all users we must also respect the contribution and the rights of the community this is an interesting question in itself do you want to have a vote-based decision making who decides about this who is the boss in the open source Community for this particular development you see a lot of interesting question I hope I got you a little bit interested in this topic maybe also from a point of innovation Theory and how to optimize the future open source community that we are all according I see thanks for watching thanks for listening I hope it was a little bit interesting and I see you in my next video okay