State of GenAI in the SAP Community 09/2023
2023-10-2 19:40:56 Author: blogs.sap.com(查看原文) 阅读量:5 收藏

In my former blog “ChatGPT in SAP Community for 2023 so far” I checked out the 10 most interesting SAP Community blogs around ChatGPT, LLM and GenAI until the first half year of 2023. Maybe you recognized a lot happens in the last months.

As I just wrapped up a current state of SAP Community for a customer presentation, I want to share my new insights in this blog.

Let’s start with some basics. What is interesting is, that currently when it comes to Artificial Intelligence (AI) the focus is very much on Large Language Models (LLM). I even got questions about if LLMs and applications like ChatGPT – and maybe you have heared about OpenAI Codex or ChatGPT Code Interpreter – would replace Data Scientists and AI developers.

But AI is a huge topic and really involves a lot of very specific tasks and capabilities used in many ways today. AI comes a longe way and is much more then what everyone can use currently as ChatGPT. We should be aware of this.

Fig. 1: History of AI until rise of Transformer Architectures and application like ChatGPT

According to Gartner Generative AI is on the Peak of Inflated Expectations. And there are still many flaws and challenges in using these technologies. In an analysis by OpenAI about the GPT4 model they observed the following “safety challenges”:

Fig. 2: Safety challenges observed with GPT4

Some of the are very well know already and also described in the SAP Community in many blogs around GenAI and LLMs.

Fig. 3: Explaination of “Hallucinations” by Wikipedia

Since than a lot of things happend working on these challenges. We see research flourishing while new concepts and terms come ab trying to offer solutions and additonal value. I just describes some of them:

Fig. 4: Currently discussed terms and concepts in GenAI

In the meantime we have seen many new blogs here in SAP Community about the topics.

Before we start, two current reading recommendations:

From SAPs Yatsea Li there is the blog series “Exploring the potential of GPT in SAP ecosystem” gives you a really good overview about how to use GenAI with SAP.

Mario De Felipe has now published a 5 part blog series about how to make use of GenAI in different contexts. Recommended for everyone want to dive deeper into the topic!

SAP Community Insights

Let’s start with some numbers. I sorted some out for quality or language reasons and came out with 82 blogs about GenAI, LLM, ChatGPT and so on. As ChatGPT and the topics around started at the end of November 2022, this is not so bad.

Fig. 5: Overview SAP Community Insights form 82 blogs

As the collection is from 11th of September, September month is not complete here. I expect the number also around the level of August.

I think we can see here a trend having some peaks over the time. Bus also that February and March blogs had been the most interested. As the topic came into the public awareness at this time, I think this is explainable.

Additionally I checked which blogs are written by SAP employees – as far as this can be identified:

Fig. 6: Overview SAP Community Insights form 82 blogs – SAP employees

Let’s explore a little bit how this fits to the general hype around ChatGPT.

Fig. 7: Hype around ChatGPT in numbers

So we see see the rise of the hype until May 2023. What happend? Is the hype already over?

I don’t think so. First we see new offerings around GenAI and LLMs come up. ChatGPT is not the only one any more. Second – we got more and more commercial offerings like Azure OpenAI which shifts the interest to more differentiated topics. So I think we are still within the hype, waiting for more.

And as you know, more happens also on the SAP side.In May SAP started a partnership with Microsoft to make use of OpenAI for customers. With the announcement of Joule, a Generative AI assistent, SAP started a feasible service, integrating GenAI into their products. SAP also made a Statement of Direction using Generative AI for analytics and planning. Just some days ago the clear direction publishing a Whitepaper and showing the direction SAP wants to go.

To understand which topics are mostly interesting here, I clustered the blogs I found on the topics:

Fig. 8: SAP Community blogs by topic

So indeed coding and developer support is one of the most interesing topics, followed by Data & Analytics and SAP Integration Suite.

There are a lot of general blogs discussing possiblities and potential of the technology. For the applications I found 5 blogs directly about S/4HANA and furthermore some blogs about different SAP cloud applications.

Let’s have a look on topics related to developers:

Fig. 9: ChatGPT & Co. for Developers

As we are in IT and development is part of every project and SAP system, this is maybe of general interest here. We can find general support for coding in ABAP, JavaScript or Python. But finaly development is the glue to get LLMs and SAP applications together.

But maybe if ChatGPT & Co. can already do so much within development – will it replace devopers in the next years? I wouldn’t expect that. We currently see a push in the way developers work. But it is a progress we always have. Sometimes faster, sometimes slower.

Please read also “Are we ready to use AI for ABAP programming?

Some additional insights why possibly developers and other jobs will not be replaced are:

  • Outdated models
  • Quality of answers and hallucinations
  • Understanding of input
  • No access to extensive ABAP/domain examples

We shouldn’t forget, that LLMs typically are general knowledge models. But we also see more domain specific models coming up and we see progress with topics like fine-tuning and the above mentioned Retrival Augmented Generation (RAG).

Therefore it is important to stay up to date and make use of new technologies in a proactive way.

SAP is offering a Reference Architecture for integration of OpenAI:

Fig. 10: Reference architecture of an SAP BTP CAP application using GPT Models of OpenAI

I put together the most relevant information I could find to get startet:

Referenz Architecture 1

  • GitHub
  • SAP AI Core as proxy

Reference Architecture 2

  • GitHub
  • CAP API consumption

Content Package for LLM 

  • Python – PyPi
  • Set up SAP AI Core
  • Incl. vector database setup

Set up Vector Database

  • Based on PostgreSQL
  • pgvector Extension

SAP Mission for Generative AI on BTP

  • Azure OpenAI service
  • Synthetic data through ChatGPT
  • Stable Diffusion for Text-to-Image
  • GitHub

So there are already many possibilites to get started, learn and make use of Generative AI with SAP.

For sure there is much more to say. I recommend to stay tuned to new blogs and updates within the SAP Community and SAP News.

I just want to close with some recommendations:

Fig. 11: General recommendations for the use of GenAI

What do you think about GenAI, LLM, ChatGPT, …? What is your current experience? How do you use these technologies together with SAP?


文章来源: https://blogs.sap.com/2023/10/02/state-of-genai-in-the-sap-community-09-2023/
如有侵权请联系:admin#unsafe.sh