Exclusive Coverage: theCUBE at IBM Think | Michaele Rosen & Keri Olson

theCUBE.net

Clips
News
More from Exclusive Coverage: theCUBE at IBM Think

Keri Olson

VP, Product Management, AI for Code

IBM

Michele Rosen

Analyst

IDC

Gen AI for coding: The expanding use cases for assisted software development

Barring the growing pains of hallucination and bias, generative artificial intelligence excels at repetitive, information-based tasks.With products such as GitHub Copilot and Amazon CodeWhisperer generally available, its capabilities are changing how developers code. How is IBM Corp. expanding the AI for coding idea from simple assistants to covering numerous facets of the software development lifecycle?IDC’s Michele Rosen (left) and IBM’s Keri Olson (middle) discuss AI for coding with theCUBE

play_circle_outline Use cases of AI for coding in software development

play_circle_outline IBM's Forward-thinking Approach: Unveiling the Adoption of Code Assistants and their Strategic Emphasis on AI

play_circle_outline Benefits of AI for code in terms of productivity and skills gap

play_circle_outline Announcement of watsonx Code Assistant for Enterprise Java Applications

Info

Michaele Rosen & Keri Olson

Keri Olson

VP, Product Management, AI for Code IBM

Michele Rosen

Analyst IDC

search

>> Welcome back to IBM Think 2024. We're here in Boston in the Seaport. Wow, what changes have occurred in the Seaport since theCUBE first started here in May of 2010? My name is Dave Vellante, and you're watching theCUBE. Keri Olson is here. She's the vice president of product management, AI for code at IBM. And Michele Rosen, she's an analyst at IDC, covering open gen AI research, a newly-minted title at IDC. Michele, you're really in all the hot spots. Keri, thanks for coming on.

Keri Olson

>> Thank you. It's good to be here.

Michele Rosen

>> Thanks for having us.

>> Quite a show. Let's start off, Michele, with the research angle. Big trends you're seeing in AI for coding, it's the hot topic of the day.

Michele Rosen

>> Well, what's interesting is how quickly gen AI is moving and how many use cases that we're seeing emerging for gen AI, even within software development. So, we've started out with coding assistants and they're generating code, but we're actually finding that developers are really interested in using gen AI technology for other aspects of the software development life cycle, such as testing or threat analysis. So, IDC has identified about 25 different gen AI use cases just within the software development life cycle. So, it's a very quick-moving sector.

>> It is moving fast, Keri, isn't it? I mean, Michele's mentioning these specific use cases. You think about testing, there's been a lot of software built up and processes and infrastructure, and a lot of that's going to get disrupted. What are you seeing in your client base in terms of the adoption of code assistants moving really fast, and what's IBM's angle there?

Keri Olson

>> Absolutely. So, AI for code is a very fast-moving space. We're very excited about it at IBM. Now, our AI for code strategy is really multifold. Number one, we are building our own Granite foundation models for code. So, these models, they are built on open source code and we have various models that range in size from 3 billion parameters to 34 billion parameters, trained on 116 programming languages. And we're very happy to share that these models for code are performing well above their weight in terms of performance and accuracy going to our users. So, that's the first thing that we're focused on. Now, you may have heard that we have recently released these code models to the open source community, and that is very important for us because we want to make coding as easy as possible for as many programmers as possible. So, that's the other thing that we're focused on. Now, from a product perspective, what are we doing there? IBM is focused on our coding assistants. So, we have a portfolio of products called IBM watsonx Code Assistant, and there we are taking those Granite foundation models for code. We are further training and fine-tuning them to focus on very specific use cases in specific domains. So, some of the domains that we're focusing on with our customers, is IT automation and application modernization, and I'll be happy to talk a little bit more about that.

>> Fascinating topics. Well, we definitely want to come back to that. When I graduated from college, one of my first jobs was coding and I wasn't a great coder, but they had me do debugging, which was the absolute worst. I hated it and left the field. How are you attacking that problem, taking away some of those mundane tasks? I presume that's something that is being welcomed with open arms.

Keri Olson

>> That's absolutely right. So, to start with, again, IBM is really focusing on end-to-end solutions that provide organizations the ability to use AI for code for specific use cases. So, one of those areas is application modernization. Now, we have watsonx Code Assistant for Z that's been available since October of last year, and that's very much focused on the end-to-end application lifecycle on the mainframe. And so, in that solution, we provide capabilities that help organizations to understand the applications that they have, not just the application itself, but the landscape that it's in, the runtime, the dependencies. We help organizations then to refactor those applications and to selectively transform applications from COBOL to Java. Now a very important part of this, and this kind of gets into what you're talking about with the mundane work that has to be done, you have to be able to test those applications that you've generated, whether it be generating them from natural language or generating them from one programming language to another. Being able to develop test cases is something that's very important, so that kind of falls in that area of let us take that work for you, make it very simple and let you focus on the higher innovation areas.

>> Right, thank you that for that. Michele, the survey research that I've seen in terms of the use cases for AI, gen AI specifically, I call them kind of chatty. So, there's obviously code assistant, there's writing marketing copy, the things that you would expect that you would get out of a ChatGPT user, but code assistants is probably the number one or number two use case. Is that what you're seeing in IDC?

Michele Rosen

>> Absolutely. Our survey data shows that software development and design is likely to be one of the most heavily-impacted areas in the first 18 months. So, by the end of 2024, we're seeing a lot of adoption, a lot of testing in this space, and I think that that's where we're going to see a lot of the first production implementations of gen AI.

>> So, what are you seeing both of you as sort of the immediate impact? Obviously, developer productivity, that's a nice hit. Where does that go from here and what are you seeing in terms of real ROI from customers?

Keri Olson

>> Absolutely. So, developer productivity is absolutely the number one thing. Also, helping with the skills gaps. So, there are a lot of skills gaps out there, regardless of what language you're talking about, but especially some of those legacy languages like COBOL, and then just helping with overall agility. So, we have a number of customers that have been taking advantage of IBM watsonx Code Assistant. Let me give you one example. Water Corporation is using IBM watsonx Code Assistant for Red Hat Ansible Lightspeed. And they're using that to help them to manage their infrastructure, using automation to manage cloud infrastructure. And they have seen significant benefits already working with IBM using watsonx Code assistant, they have estimated that they are going to save over 1,500 manual hours every year by using this automation and the capability that we're offering from generative AI.

>> Nice. Okay. So, that's music to the CFOs here. How about this idea of domains specificity? theCUBE put out their gen AI power law last year, and we predicted that, and it's starting to happen, but looking for tangible examples of industry-specific or domain-specific, maybe not LLMs, maybe smaller language models. Are you seeing that and where does code assistant fit in?

Michele Rosen

>> Yeah, purpose-built is really the word for enterprise use of gen AI. I mean because you need to incorporate your enterprise data to ground the results that are coming from the LLM, you need to incorporate your enterprise data to make sure that you're communicating in the way that fits your corporate culture, using the right terminology, the right policies and background data. So, we absolutely see a development of, I've seen people call the medium language models or small language models, but smaller, let's say, than the largest sizes of language models. And because that's really sort of the ideal of the cost performance point, where they're able to take this base model and then train it using their own data, using a couple of different methodologies that I think Keri might talk to you about, and then really get much better results. I think that's the key point is when you see developers going out to a general purpose coding assistant, they're not likely to get the best results, and sometimes that can be a little disheartening. So, you really need to step back and look at your corporate code base and then take advantage of that to get better results from these tools.

>> So, speaking of methodologies, can you talk to your specific purpose-built approach? What's unique about it from IBM's perspective?

Keri Olson

>> Absolutely. So, we do believe in a purpose-built approach and having targeted solutions for specific domains. So, I've already talked about our large language models, the fact that we have trained them on 116 programming languages, that they're now available in the open source community. But what we do in IBM is we then take them and further train them and fine tune them for these specific domains. And with that, as Michele mentioned, you have smaller models that are purpose trained for very specific use cases, and we are seeing greater results, greater performance, and greater accuracy based on these models. But it doesn't stop at the model. Our approach is to bring an end-to-end solution together combining automation with generative AI to truly provide those software development capabilities that our customers are looking for.

>> So, the smaller model makes it more manageable, I understand that. But I would think that a lot of the equation is the better data, more focused data, the data that you trust, maybe it's inherently not biased. So, can you talk to the data angle here?

Keri Olson

>> Absolutely. So, data is key when it comes to generative AI, as you've already mentioned. So, the fact of the matter is that you start with a model and then you further train that and fine tune that on the data that you have with the expertise that you have in-house. So, we take our knowledge and our skills and infuse that into the generative AI models to make them better performing. So, it's not just about the fact that it's smaller, but it's smaller and it's focused.

>> Can you walk us through the announcements of this week, just so the audience... Can you summarize that, like the quick hits? What was the hard news in terms of product that you're actually announcing?

Keri Olson

>> Yes. So, a few things that we are focused on is in the AI for code space, number one is the fact that we now have made our models available in open source. So, our Granite code models are available to the open source community. And what we are doing from a product perspective is that we are continuing to build on the products that we have and the capabilities that we have in our AI for code space. So, watsonx code assistant for Z and watsonx code assistant for Red Hat Ansible Lightspeed, they were both launched in October of last year. Now, we continue to build on those products, bringing new capabilities. So, some of the key capabilities coming into watsonx Code Assistant for Z, which is really focused on the end-to-end application life cycle on the mainframe is COBOL explanation. So, explanation is one of the key capabilities that we're going to-

>> What's this code do?

Keri Olson

>> Exactly right. So, what does this code actually do? Explain this to me. That is one thing that we have just announced. We have also announced COBOL optimization. We are also announcing that our products, both watsonx Code Assistant for Z and watsonx Code Assistant for Red Hat Ansible Lightspeed, will be available on premises in addition to being SaaS. So, now we're providing hybrid cloud capabilities to our customers. And this is really a game-changer for those organizations who do not want their data to lead their premises. They can put all of this in their own data center, and utilize it just like customers are today on the cloud. Now, one thing that I'm really excited about is that we are announcing a new product. We just announced this yesterday, IBM watsonx Code Assistant for Enterprise Java Applications. And this product, again, is a domain-specific product that's focused on accelerating the application lifecycle for enterprise Java applications. We will be helping with Java upgrades, Java enhancements, Java modernization. So, you can modernize your runtime from, say, traditional WebSphere to Liberty. So, these are some of the key capabilities that will be coming in watsonx Code Assistant for Enterprise Java Applications. It is going to be in private preview in June, and we expect it to launch as a generally-available product by the end of the year.

>> Wow, okay. A lot of Java out there, so that's key.

Keri Olson

>> Yes.

>> And my understanding is that, and I haven't dug into all the licenses. I've definitely read the fine print on Llama and Llama 3. My understanding is that IBM's licenses are the least restrictive, I'll say it that way. And I think there's indemnification involved. I don't know if it flies into or bleeds into code assistant. Can you explain that, please, your point of view on this?

Keri Olson

>> Yeah, so in terms of the licensing of the models themselves, so the models that we have just released to open source are released under an Apache 2 license. So, that is a very open license. It's available for users and organizations to use as they see fit. We do offer indemnification with our code models as well, and that does of course extend to our end-to-end solutions around watsonx Code Assistant.

>> Michele, you're seeing certain language in some of these models. It says, "Well, if you're really big, we could pull the plug and we might charge you..." You guys aren't going there, I'm inferring.

Michele Rosen

>> Yeah, open source is complicated in the context of language models because it's not just code. There's code, there's the data, then there are use policies that are being put in place. So, as you mentioned for Llama 3, you have to sign a responsible use policy. For some of the other open models, you have to sign responsible use policies. And while you can understand that because people have obvious concerns about risks related to gen AI, there's also a really strong ethic around the open source community and the idea that with enough eyeballs, all bugs are shallow. And we really see an opportunity for open source models to engender a level of trust that might not be possible for developers and organizations to develop with a closed source model. They just don't know what's happening on the inside.

>> Yeah, I've talked to customers who said they're really concerned about that. And they're actually going to build, believe it or not, their own language models. We'll see if that is a trend that survives the long term. Last question. I'll start with Michele, and then Keri, you can bring us home. One year from now, we're at Think 2025, what do you want to be able to say that you can't say today?

Michele Rosen

>> That's a really good question. I think that the hardest challenge right now is, again, with the number of use cases that are out there and the kind of unique aspects of every organization's data and IT infrastructure, it's really difficult to find a packaged solution that will meet everyone's needs. So, everybody's kind of experimenting. We need the help of consultants and other assistants. We need to decide on the use cases that are going to be of most value to the company. But I'm really hoping that by next year we can maybe say, "Okay, you want to fine-tune your model? Here's the methodology. Here are the products that are available." But only a few companies like IBM have really released end-to-end platforms at this point.

>> And Keri, how about it? What do you want to be able to say a year from now that you can't say today?

Keri Olson

>> So, I really agree with Michele that there is a strong need, not only for code assistants that come out of the box with the capabilities, the languages, et cetera, that organizations are looking for, but giving organizations the ability to customize that coding assistant based on their individual needs, I think is extremely important. So, we have just announced InstructLab, right? In addition to Open-sourcing the models, we now have InstructLab. And the beauty of InstructLab is it will allow organizations to bring in their own skills and their own knowledge to further train the model for their needs. And that can then be applied to the code assistants as well. So, as organizations start using watsonx Code Assistant, my goal in the next year would be to say, "And you have the ability to use InstructLab to bring in your own skills and knowledge to make this your own, whatever that may be. If you have a niche programming language, if you have a legacy programming language, whatever you need to use that coding assistant for, that's where IBM wants to be, helping you to be able to customize that and make that your own, such that you're getting the absolute most value you can out of IBM watsonx Code Assistant."

>> And you can build a moat because your moat is your data.

Keri Olson

>> Absolutely.

>> And my understanding is I can choose whether or not to release that into the public-

Keri Olson

>> You're absolutely right....

>> or I can keep that to myself as my proprietary advantage.

Keri Olson

>> You are right.

>> Guys, thanks so much.

Keri Olson

>> Thank you so much.

Michele Rosen

>> Thank you.

>> Really appreciate your time and best of luck.

Michele Rosen

>> Thank you.

Keri Olson

>> I appreciate it.

>> Okay. Keep it right there. We're here at IBM Think 2024. You're watching theCUBE. My name is Dave Vellante. We'll be right back, right after this short break.

Michaele Rosen & Keri Olson

search

>> Welcome back to IBM Think 2024. We're here in Boston in the Seaport. Wow, what changes have occurred in the Seaport since theCUBE first started here in May of 2010? My name is Dave Vellante, and you're watching theCUBE. Keri Olson is here. She's the vice president of product management, AI for code at IBM. And Michele Rosen, she's an analyst at IDC, covering open gen AI research, a newly-minted title at IDC. Michele, you're really in all the hot spots. Keri, thanks for coming on.

Keri Olson

>> Thank you. It's good to be here.

Michele Rosen

>> Thanks for having us.

>> Quite a show. Let's start off, Michele, with the research angle. Big trends you're seeing in AI for coding, it's the hot topic of the day.

Michele Rosen

>> Well, what's interesting is how quickly gen AI is moving and how many use cases that we're seeing emerging for gen AI, even within software development. So, we've started out with coding assistants and they're generating code, but we're actually finding that developers are really interested in using gen AI technology for other aspects of the software development life cycle, such as testing or threat analysis. So, IDC has identified about 25 different gen AI use cases just within the software development life cycle. So, it's a very quick-moving sector.

>> It is moving fast, Keri, isn't it? I mean, Michele's mentioning these specific use cases. You think about testing, there's been a lot of software built up and processes and infrastructure, and a lot of that's going to get disrupted. What are you seeing in your client base in terms of the adoption of code assistants moving really fast, and what's IBM's angle there?

Keri Olson

>> Absolutely. So, AI for code is a very fast-moving space. We're very excited about it at IBM. Now, our AI for code strategy is really multifold. Number one, we are building our own Granite foundation models for code. So, these models, they are built on open source code and we have various models that range in size from 3 billion parameters to 34 billion parameters, trained on 116 programming languages. And we're very happy to share that these models for code are performing well above their weight in terms of performance and accuracy going to our users. So, that's the first thing that we're focused on. Now, you may have heard that we have recently released these code models to the open source community, and that is very important for us because we want to make coding as easy as possible for as many programmers as possible. So, that's the other thing that we're focused on. Now, from a product perspective, what are we doing there? IBM is focused on our coding assistants. So, we have a portfolio of products called IBM watsonx Code Assistant, and there we are taking those Granite foundation models for code. We are further training and fine-tuning them to focus on very specific use cases in specific domains. So, some of the domains that we're focusing on with our customers, is IT automation and application modernization, and I'll be happy to talk a little bit more about that.

>> Fascinating topics. Well, we definitely want to come back to that. When I graduated from college, one of my first jobs was coding and I wasn't a great coder, but they had me do debugging, which was the absolute worst. I hated it and left the field. How are you attacking that problem, taking away some of those mundane tasks? I presume that's something that is being welcomed with open arms.

Keri Olson

>> That's absolutely right. So, to start with, again, IBM is really focusing on end-to-end solutions that provide organizations the ability to use AI for code for specific use cases. So, one of those areas is application modernization. Now, we have watsonx Code Assistant for Z that's been available since October of last year, and that's very much focused on the end-to-end application lifecycle on the mainframe. And so, in that solution, we provide capabilities that help organizations to understand the applications that they have, not just the application itself, but the landscape that it's in, the runtime, the dependencies. We help organizations then to refactor those applications and to selectively transform applications from COBOL to Java. Now a very important part of this, and this kind of gets into what you're talking about with the mundane work that has to be done, you have to be able to test those applications that you've generated, whether it be generating them from natural language or generating them from one programming language to another. Being able to develop test cases is something that's very important, so that kind of falls in that area of let us take that work for you, make it very simple and let you focus on the higher innovation areas.

>> Right, thank you that for that. Michele, the survey research that I've seen in terms of the use cases for AI, gen AI specifically, I call them kind of chatty. So, there's obviously code assistant, there's writing marketing copy, the things that you would expect that you would get out of a ChatGPT user, but code assistants is probably the number one or number two use case. Is that what you're seeing in IDC?

Michele Rosen

>> Absolutely. Our survey data shows that software development and design is likely to be one of the most heavily-impacted areas in the first 18 months. So, by the end of 2024, we're seeing a lot of adoption, a lot of testing in this space, and I think that that's where we're going to see a lot of the first production implementations of gen AI.

>> So, what are you seeing both of you as sort of the immediate impact? Obviously, developer productivity, that's a nice hit. Where does that go from here and what are you seeing in terms of real ROI from customers?

Keri Olson

>> Absolutely. So, developer productivity is absolutely the number one thing. Also, helping with the skills gaps. So, there are a lot of skills gaps out there, regardless of what language you're talking about, but especially some of those legacy languages like COBOL, and then just helping with overall agility. So, we have a number of customers that have been taking advantage of IBM watsonx Code Assistant. Let me give you one example. Water Corporation is using IBM watsonx Code Assistant for Red Hat Ansible Lightspeed. And they're using that to help them to manage their infrastructure, using automation to manage cloud infrastructure. And they have seen significant benefits already working with IBM using watsonx Code assistant, they have estimated that they are going to save over 1,500 manual hours every year by using this automation and the capability that we're offering from generative AI.

>> Nice. Okay. So, that's music to the CFOs here. How about this idea of domains specificity? theCUBE put out their gen AI power law last year, and we predicted that, and it's starting to happen, but looking for tangible examples of industry-specific or domain-specific, maybe not LLMs, maybe smaller language models. Are you seeing that and where does code assistant fit in?

Michele Rosen

>> Yeah, purpose-built is really the word for enterprise use of gen AI. I mean because you need to incorporate your enterprise data to ground the results that are coming from the LLM, you need to incorporate your enterprise data to make sure that you're communicating in the way that fits your corporate culture, using the right terminology, the right policies and background data. So, we absolutely see a development of, I've seen people call the medium language models or small language models, but smaller, let's say, than the largest sizes of language models. And because that's really sort of the ideal of the cost performance point, where they're able to take this base model and then train it using their own data, using a couple of different methodologies that I think Keri might talk to you about, and then really get much better results. I think that's the key point is when you see developers going out to a general purpose coding assistant, they're not likely to get the best results, and sometimes that can be a little disheartening. So, you really need to step back and look at your corporate code base and then take advantage of that to get better results from these tools.

>> So, speaking of methodologies, can you talk to your specific purpose-built approach? What's unique about it from IBM's perspective?

Keri Olson

>> Absolutely. So, we do believe in a purpose-built approach and having targeted solutions for specific domains. So, I've already talked about our large language models, the fact that we have trained them on 116 programming languages, that they're now available in the open source community. But what we do in IBM is we then take them and further train them and fine tune them for these specific domains. And with that, as Michele mentioned, you have smaller models that are purpose trained for very specific use cases, and we are seeing greater results, greater performance, and greater accuracy based on these models. But it doesn't stop at the model. Our approach is to bring an end-to-end solution together combining automation with generative AI to truly provide those software development capabilities that our customers are looking for.

>> So, the smaller model makes it more manageable, I understand that. But I would think that a lot of the equation is the better data, more focused data, the data that you trust, maybe it's inherently not biased. So, can you talk to the data angle here?

Keri Olson

>> Absolutely. So, data is key when it comes to generative AI, as you've already mentioned. So, the fact of the matter is that you start with a model and then you further train that and fine tune that on the data that you have with the expertise that you have in-house. So, we take our knowledge and our skills and infuse that into the generative AI models to make them better performing. So, it's not just about the fact that it's smaller, but it's smaller and it's focused.

>> Can you walk us through the announcements of this week, just so the audience... Can you summarize that, like the quick hits? What was the hard news in terms of product that you're actually announcing?

Keri Olson

>> Yes. So, a few things that we are focused on is in the AI for code space, number one is the fact that we now have made our models available in open source. So, our Granite code models are available to the open source community. And what we are doing from a product perspective is that we are continuing to build on the products that we have and the capabilities that we have in our AI for code space. So, watsonx code assistant for Z and watsonx code assistant for Red Hat Ansible Lightspeed, they were both launched in October of last year. Now, we continue to build on those products, bringing new capabilities. So, some of the key capabilities coming into watsonx Code Assistant for Z, which is really focused on the end-to-end application life cycle on the mainframe is COBOL explanation. So, explanation is one of the key capabilities that we're going to-

>> What's this code do?

Keri Olson

>> Exactly right. So, what does this code actually do? Explain this to me. That is one thing that we have just announced. We have also announced COBOL optimization. We are also announcing that our products, both watsonx Code Assistant for Z and watsonx Code Assistant for Red Hat Ansible Lightspeed, will be available on premises in addition to being SaaS. So, now we're providing hybrid cloud capabilities to our customers. And this is really a game-changer for those organizations who do not want their data to lead their premises. They can put all of this in their own data center, and utilize it just like customers are today on the cloud. Now, one thing that I'm really excited about is that we are announcing a new product. We just announced this yesterday, IBM watsonx Code Assistant for Enterprise Java Applications. And this product, again, is a domain-specific product that's focused on accelerating the application lifecycle for enterprise Java applications. We will be helping with Java upgrades, Java enhancements, Java modernization. So, you can modernize your runtime from, say, traditional WebSphere to Liberty. So, these are some of the key capabilities that will be coming in watsonx Code Assistant for Enterprise Java Applications. It is going to be in private preview in June, and we expect it to launch as a generally-available product by the end of the year.

>> Wow, okay. A lot of Java out there, so that's key.

Keri Olson

>> Yes.

>> And my understanding is that, and I haven't dug into all the licenses. I've definitely read the fine print on Llama and Llama 3. My understanding is that IBM's licenses are the least restrictive, I'll say it that way. And I think there's indemnification involved. I don't know if it flies into or bleeds into code assistant. Can you explain that, please, your point of view on this?

Keri Olson

>> Yeah, so in terms of the licensing of the models themselves, so the models that we have just released to open source are released under an Apache 2 license. So, that is a very open license. It's available for users and organizations to use as they see fit. We do offer indemnification with our code models as well, and that does of course extend to our end-to-end solutions around watsonx Code Assistant.

>> Michele, you're seeing certain language in some of these models. It says, "Well, if you're really big, we could pull the plug and we might charge you..." You guys aren't going there, I'm inferring.

Michele Rosen

>> Yeah, open source is complicated in the context of language models because it's not just code. There's code, there's the data, then there are use policies that are being put in place. So, as you mentioned for Llama 3, you have to sign a responsible use policy. For some of the other open models, you have to sign responsible use policies. And while you can understand that because people have obvious concerns about risks related to gen AI, there's also a really strong ethic around the open source community and the idea that with enough eyeballs, all bugs are shallow. And we really see an opportunity for open source models to engender a level of trust that might not be possible for developers and organizations to develop with a closed source model. They just don't know what's happening on the inside.

>> Yeah, I've talked to customers who said they're really concerned about that. And they're actually going to build, believe it or not, their own language models. We'll see if that is a trend that survives the long term. Last question. I'll start with Michele, and then Keri, you can bring us home. One year from now, we're at Think 2025, what do you want to be able to say that you can't say today?

Michele Rosen

>> That's a really good question. I think that the hardest challenge right now is, again, with the number of use cases that are out there and the kind of unique aspects of every organization's data and IT infrastructure, it's really difficult to find a packaged solution that will meet everyone's needs. So, everybody's kind of experimenting. We need the help of consultants and other assistants. We need to decide on the use cases that are going to be of most value to the company. But I'm really hoping that by next year we can maybe say, "Okay, you want to fine-tune your model? Here's the methodology. Here are the products that are available." But only a few companies like IBM have really released end-to-end platforms at this point.

>> And Keri, how about it? What do you want to be able to say a year from now that you can't say today?

Keri Olson

>> So, I really agree with Michele that there is a strong need, not only for code assistants that come out of the box with the capabilities, the languages, et cetera, that organizations are looking for, but giving organizations the ability to customize that coding assistant based on their individual needs, I think is extremely important. So, we have just announced InstructLab, right? In addition to Open-sourcing the models, we now have InstructLab. And the beauty of InstructLab is it will allow organizations to bring in their own skills and their own knowledge to further train the model for their needs. And that can then be applied to the code assistants as well. So, as organizations start using watsonx Code Assistant, my goal in the next year would be to say, "And you have the ability to use InstructLab to bring in your own skills and knowledge to make this your own, whatever that may be. If you have a niche programming language, if you have a legacy programming language, whatever you need to use that coding assistant for, that's where IBM wants to be, helping you to be able to customize that and make that your own, such that you're getting the absolute most value you can out of IBM watsonx Code Assistant."

>> And you can build a moat because your moat is your data.

Keri Olson

>> Absolutely.

>> And my understanding is I can choose whether or not to release that into the public-

Keri Olson

>> You're absolutely right....

>> or I can keep that to myself as my proprietary advantage.

Keri Olson

>> You are right.

>> Guys, thanks so much.

Keri Olson

>> Thank you so much.

Michele Rosen

>> Thank you.

>> Really appreciate your time and best of luck.

Michele Rosen

>> Thank you.

Keri Olson

>> I appreciate it.

>> Okay. Keep it right there. We're here at IBM Think 2024. You're watching theCUBE. My name is Dave Vellante. We'll be right back, right after this short break.