AI and the Future of Search
内容介绍：
一、The current state and challenges of search technology
二、Product and technology development at Elastic
三、Future outlook and resources
本次分享的主题是AI and the Future of Search，由Elastic, Global Vice PresidentBahaaldine Azarmi分享。
Thanks for the transition called，I am Bahaaldine Azarmi，an Elastic，Global Vice President。I am very pleased to be here today and very honored to get a chance to talk about the last weekend of today.
一、The current state and challenges of search technology
1.We 've Come a Long Way
So what we're going to talk about this, just search. There's a lot of display，but I wanna tell you is that we came up a long way.
We started with you the community for the developers、involved on solution、 started with political search，a lot of contribution in the solution. And then we have all the solution to a modern politics search. We added applications、 search、filtering also for primitive that you need to build a great sort of experience. And for us, it's very important that we bring to the builders, to the developers, what they need to be able to search experiences.
2.Search has changed how the world uses data
One of the things that we see is our customers completely changed, the way they use their data from search. Those numbers doesn't like they tell you the story that a lot of people are downloading elastic, i'm sure most of you know about the last search and some of you have built solution with the last search. This is really the result of that word between us in the community.
Our engineering team has been working really hard to make sure that you get the air capability. You need a solution to build great, genuine experiences. We made our product to involve our solution involved as well. Now we define ourselves as the surgery company because we believe that search is one of the most important aspect of AI.
3.Search is more important than ever
This is something we started to believe the first eleven went up, but three years ago, when we started to build our factor database capabilities. Without this is gonna a great competition for AI、forspecific reason.
We felt that even if you train the model on the large context and having a knowledge that makes you generalist, you will never get accuracy for something that is private data that is a so when you combine analysis with search is between weather as a search, if you that combination where you ground the airline to acknowledge that is more accurate.
4.Generative AI comes with specific challenges
Think about eliminate models to us. When we think about this and we think about this very high level, we think about this is the new provider of the data. It's a new way to interact with the data. It gets into your application and I can tell the data sources. So It comes with great benefit. Don't get me wrong. It's awesome everybody wants to reveal the application with the land today, but what there is couple of downsides.
The first one that pretty much everybody knows about the intelligence nation and is the fact of not having something that is accurate in terms of the answer. The answer versus getting something that is more accurate than the way which you can safely take a decision because at the end of the day, those data that the natural name is bringing you are you genuinely application is bringing to the user and to take a decision. So how this nation can be very bad.
The other thing is I combine this in two buckets, but in the privacy and making sure that you understand who is accessing the data, but that the grand narrative that is fine enough. So we have a full control of what in the last few years. But the other aspect is not only the data privacy, but it's the real time access to a private data and essentially taking a concept of data that is private and bringing into the Ellen.
二、Product and technology development at Elastic
1.Retrieval Augmented Generation
It's all the way architecture thinks about IT is taking elastic as a center piece of the architecture where they are bringing private data in terroristic. They connected to their headlines. And instead of asking to the editor in the answer or generating data.
She research blue and I think first, and they submit, the results to uh and then get something that is like great. So this is what we call a rag.
2.Builiding with GAI：spectrum of approaches
You are talking about enough of that alibaba is doing your this is a typical architecture, but there is no other solutions. It's not just about right. One of the things you can do or two of the things you can do is pretty in the model or find in the model. It's easier said than actually done. There is couple of also downside to eats. It's possible, but in percent expertise is not something that is straight forward and everybody can do.
It takes time,You requires data and require some sort of tagging method work and also training and bitch working in evaluation. And the oil speaking is IT could be expensive. It's not something that is cheap. The way our customers are thinking about Richard war management generation is the shortest path to value when IT comes to leveraging a natural, and it's not the most perfect.
3.A culmination of years of investment
We will last week,It doesn't lie as well when our customers started to use elastic or dogs, we will build a love of the ability when we start seeing people building a last week using a sick installed in security Operation center to do thread hunting, we will be the security solution.
4.Elasticsearch Relevance Engine
Now we are seeing our consumers using IT for Jenny eye. We'll be initially a lastic search relevance engine is the set of tools that we are bringing to builders. Because there are important in the context of Jenny eye extra similarity pictures. Being able to choose a model, bring your model, generate your embarrasses, changing semantic, semantic search.
And all of that is part of israel. And enterprise reading features, where you don't feel like you're just trying to a solution to Justin. What you actually put the you can put IT in production. And we have a lot of customers already in production with.
5.Make Elasticsearch the BEST open source Vector Database
When I talk about those future that are helping in the general economics, think about this is the future that you will use for Victoria days, mainly, but also for search. And what we intend to do with our solution is to make that I think, search the first open source of the database in the world. That is a goal。
6.Only Elastic provides ALL the capabilities you need
I just wanted to put this in perspective when we say, when we say Victor database .Victor's similarity is probably in the in the red circle. Maybe the database in that blue circle. And then as you, as you build the search experience. You go to a on the search, application, whatever website you use you go on that search box. There are different ways you are searching.
You don't always asking a question in the search box. You're borrowing good. He wore a location, something that defines something you wanted to do for. This is not something that a Victory base can give by itself only the full set of baby item will help you to build a great search. For us, Victory, Victory basis is not a separate product. It's not a separate module. It's just another set of API. We added into a lesson search.It's part of the ecosystem of pictures. You already know, you should be very familiar with using those users. If you build the solution, we will assist before.
7.Make Elastic the most OPEN member of the Gen AI ecpsystem
One of the things also that our engineering team is working hard . It is not only bringing you the best pictures to create the basic GI experience, we want to make sure that we stay relevant to the markets in terms of the world. We are bargaining with from a technology perspective. Like launching lawyer and go here.
For example. On the revenging API,We are using the cohered model. Because we believe that, that's the. One of the most relevant in the market.
For now. Those are the technology partnership where building still uh every day. Actually, we think about what are the technology we want to integrate with and particularly to bring those experience to our developers.
三、Future outlook and resources
1.What's Next for Search?
What's Next for Search?If I if I had to summarize, very simply good, we want to make genuine easy, and make IT relevance. It's easy to create the genuine application when you just block a natural name. It doesn't mean it's gonna be relevant. It's not necessarily easy for anyone to start to prevent your application. We want to reduce the barrier of entry by abstracting. The features you need to understand to create those application.
2.Make Elasticsearch the BEST at Semantic Search and Retrieval Augmented Generation
We're pretty uh happy with the evolution we're seeing right now and the results have there is a lot of work in general team is putting into optimizing our victims of these capabilities. If you take, for example, scale acquaintances as the effect of reducing the footprint of evidence, you could in memory. This is something we've been working on a lot.
We worked on the info quantization lately. Reducing by four times the size of embedding in the memory. We are not working on binary. So we want to reduce IT even more. And this is something we are making for you, the developers. So you can be on application.
We even more vectors and embedding in memory and having something even more responsive and agreed in terms of semantic search.
3.Our Search product team Priorities
I want to talk about the priorities for our search team for our it's basically for the search. One of the priority we have is to build an influence service that will be a first last service. This influence surveys will allow any developer to connect to different analysis and will bring features for governing Operation on top of the zealand's. This is something that's going to be that in the worst is going to be available sometime
this is very important for us. Because we see that the volunteer is not necessarily want to integrate this, specifically allowed to their application. They wanted something that attracted this and only that they want a service between their application and the element of the management. We are focusing on a laughter is on boarding. What we want is for a developer to come to electric search get API Key, get API and start pushing for employees in the solution.
This is what we call that they zero experience. You are in the train, you have your left off, and you have a couple of documents that you wanna just try elastic. We are focusing on these moments where you don't know the analysis or you go on any cloud, and you want to try uh elastic. This is something we can reduce much more. And we we want to give you the experience that you can just sit up in the minutes.
So that's the Young body. The other thing on this slide is the genuine i've done through that is interesting for you to know. In the last week , you want to build before that gives its available if you wanted to build a right application, you have to come to figure out of everything. In terms of what, what do you want to be at the integration with an announcement the query that is behind the right application. All of and so we do something called a playground. So playground is in kibo Na RUI.
You going to bond up, and you have this your eyes very easy. You just set up your own name and then sit, you regard recommendation is done. What's interesting about this is you can. As many industries as you want. If you have industries in the last six weeks, you take IT and you connected to uh to this regulation. Normally you have integration with the analysis, but you ground the airline to acknowledge that is contained in the index. To get you started very easily.
And you can do that and tested in two keyboard. But the other aspect you can do is just generate the code photo. You take the code and go to point the application, so that's something we will also bring a lot of football on that cookies. And the other I speak to see on the slide, and i'm going to talk about this also in the next one is the search features we are putting together like RF to the hybrid research or learn to range or the ranking.
All goods are, not something we're just bringing to make It more complex. It's hoping in the process of accuracy to remain, for example. The result that are coming up. Aquarium and make sure that we exposing to your user and to the editor the most accurate response before generating anything so uh, this is hoping to get even more relevant contents in your argue application.
So 2024 is a pretty good year for us. So far, we had a lot of development going on. There is one that I wanna talk about this yesterday. You haven't heard about this. It's elastic search query language. And so one thing we wanted and asked if we wanna we want you to stop learning Jason. If you know what I mean about that, I think certain or crazy as early issues. It's a big Jason language. It's not always easy. Sometimes you can be secretary. That could be a drama. you should have a democratic. Now we move to a, uh, something different. So the quality of sellers still hear you love IT.
Good for you. Just keep using IT. But is there there is something else called yesterday. And so that's a pipe, language. Instead of creating a Jason, you go and just combine different function with fights. If you have an index code, my index you can say from my index in the user, a function function that could be start average of my field just to get the average of the field containing that indexes that as easy as that is your issue, we were actually very excited about the capabilities of israel. And four democrats. We we see that as a developer search. If they wonder if they want something that uh to uh, uh in the other aspect, which is very interesting that this is really so structured in terms of language, and this part, of the common language, those, Ellen, when they learn on, when the enemies are learning on violent public knowledge. They learned about everything we do.
Because we are open everything that this is described in our. In our stack. Yesterday is part of this description. You can already take a longer today and just go and ask to generate the neighborhood gurry IT will probably know how to, the other aspect. We, we saw uh, also influenced slide is automatic chunk in.
So if you were to take a document, for example, of PDF and pretty large PDF and wanna showed into elastic to generate and buildings and start doing semantic search on IT. It's not always easy because you have to go through this process where you take the video and tranquil into multiple parts of until uh, uh, a feature we call semantic text. You have to do IT manually. Now, everything is gone for you in the background. And so the text is jumped into multiple messages. Over that at the beginning and at the end of the passengers who, when you do a semantic search on those content, there is a continuity that is built by the overwhelming of the passengers. And it's much easier for you to justify more data in two elections.
4.You can go from insight to outcome faster with search powered AI
I talked a lot about those primitive, and you saw those uh features. One of the very interesting aspect with those features。 The report of the so any solution we build on top of the platform benefit and leverage those features to define a threat hunter. I am in the security solution. And I want I can just query with the history of directly in the kitchen to search for the host where analysis ray went. We have two assistance. One for ability and one for security that are actually regulator using the same capabilities that I talk about. They literally uh, use uh elements and industries to bring interesting data.
For example, and answer in the room and you are looking for an incident in one of your host. And you shall not understand, whether somebody have seen IT uh, with in the authority before you. One thing you can ask to the assistance in our ability solution is there a tickets open for it.If you don't if they ever learn. Yeah yesterday, but it's a whole note of that. But the one thing you can do with the assistant is to bring the index that contains all the tickets in the campaign into the assistant. And it will know the conflict of the rest of the story. So those are very interesting features that all are solution or benefit.
5.RESOURCES FOR DEVELOPERS :Elastic Labs
There is a lot in what I explain.There is a way you can learn about this. We have to read that this is pretty awesome. If you want to learn, take a picture of this because there is a good thread of the electric dog,Coke, flash. I started searching out of the victim and security that we have a lot of constant on this. And those are content. We have to get started.
We have real life use case. Which is described that by step, you can take IT just go in corruption. You want. There is a lot of art we talk about new fiction that I wana talk.This one of the one of the one of the world recently is about users will, for analytics, I wrote another one of multiple agents, architecture to leverage. And then there's really a lot of content here. You will be able to leverage.
6.Elastic *Alibaba Cloud Win -Win in AI Era
I just wanna tell you that we are working, rick, with the other team, and generating team to just make sure that we are bringing you the best pictures, the one that are relevant for the chinese markets. We just had a couple of meetings today with interesting things that to fit as much as much as possible.
As big as we're working hand in hand to build the community in the world Elastic developers. This is very important for us. we commitments are to work with alibaba and ecosystem of partners. And the custom to accelerate the adoption of philosophy. Search for genuine application has never been strong. Thank you to everyone listen to me, and thank you for this opportunity to representatives.
