Swinburne University of Technology

Amir Aryani

Integrating Large Language Models (LLMs) such as GPT into organizations’ data workflows is a complex process with various challenges. These obstacles include but are not limited to technical, operational, ethical, and legal dimensions, each presenting hurdles that organisations must navigate to harness the full potential of LLMs effectively.

What are the challenges in integrating LLMs into organisations' data workflows?

I asked Generative AI Models about their context window. Their response was intriguing. The context window for a large language model (LLM) like OpenAI’s GPT refers to the maximum amount of text the model can consider at any one time when generating a response. This includes both the prompt provided by the user and the model’s generated text.

What is your Context Window?

Authors: Hui Yin, Amir Aryani  As we discussed in our previous article “A Brief Introduction to Retrieval Augmented Generation (RAG)”, RAG is an artificial intelligence framework that incorporates the latest reliable external knowledge and aims to improve the quality of responses generated by pre-trained language models (PLM). Initially, it was designed to improve the performance of knowledge-intensive NLP tasks (Lewis et al., 2020). As

Navigating the Present: Exploring Practical Horizons of Retrieval-Augmented Generation (RAG)

At the time of writing this post (November 19, 2023), there are 401,014 models and transformers listed on Huggingface.co. This number is a testament to the success and impact of the open-source model in the development of AI technologies. However, the challenge at hand is understanding the available models and keeping up to date with the latest developments.

Estimated time: 12 Years to Review all 400k Models on Hugging Face

Prompt engineering is a technique used to effectively communicate with large language models (LLM) like GPT-3 or GPT-4 and get the desired output. Here there are applications of 8 distinct prompt techniques to interact with Mistral AI and find out how we can prompt effectively to learn about Foundation Models.
<strong>
 Zero-Shot Learning
</strong>
: This involves giving the AI a task without any prior examples.

8 Types of Prompt Engineering

There is a plethora of articles and predictions about the future development of AI. Reading through them suggests three main areas of high-impact changes that would transform our daily lives: Healthcare, Automation, and Manufacturing. The impact of AI in these industries is profound and multifaceted, with each sector experiencing unique transformations. Some of these ideas are described in this NVIDIA article by Cliff Edwards.

Revolutionizing Wellness and Wheels: AI’s Transformative Role in Health and Automotive Industries

Procrastination, a common adversary in our daily lives, often leads to a backlog of tasks and ambitions. It’s a silent thief, stealthily stealing opportunities and time. You might have a stack of books on your bookshelf, longing to be read. Perhaps there’s a new language you’ve wanted to learn, or piles of paperwork cluttering your office that you’ve been meaning to organize.

Maximizing Productivity in 10-Minute Bursts

Authors: Hui Yin , Amir Aryani  With the increasing application of large language models in various scenarios, people realize that these models are not omnipotent. When generating dialogues (Shuster et al., 2021), the models often produce hallucinations, leading to inaccurate answers.

A Brief Introduction to Retrieval Augmented Generation (RAG)

Stories by Amir Aryani on Medium

<figure>
<img
src="https://cdn-images-1.medium.com/max/1024/0*VnD8bwqPJmEMEoqQ" />
<figcaption>Photo by <a
href="https://unsplash.com/@iammottakin?utm_source=medium&amp;utm_medium=referral">Mojahid
Mottakin</a> on <a
href="https://unsplash.com?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption>
</figure>

Integrating Large Language Models (LLMs) such as GPT into organizations'
data workflows is a complex process with various challenges. These
obstacles include but are not limited to technical, operational,
ethical, and legal dimensions, each presenting hurdles that
organisations must navigate to harness the full potential of LLMs
effectively. Here's an expanded categorisation:

### Technical Challenges

1.  **Data Privacy and Security:** Ensuring the security and privacy of
    data when using LLMs is a primary concern. Sensitive information
    might be exposed to unauthorized parties if proper safeguards are
    not in place.
2.  I**ntegration Complexity:** Integrating LLMs with existing data
    systems and workflows can be complex, requiring significant
    customization and configuration to ensure seamless operation.
3.  **Scalability and Performance:** Ensuring the LLM can handle the
    scale of data and provide responses promptly can be challenging,
    especially for large organizations with vast amounts of data.
4.  **Data Quality and Bias**: LLMs can perpetuate or even amplify
    biases present in their training data. Ensuring the model generates
    accurate, unbiased outputs requires constant monitoring and
    adjustment.
5.  **Model Updating and Maintenance**: Keeping the LLM updated with the
    latest information and ensuring it evolves with the organization's
    needs requires ongoing effort and resources.

### Operational Challenges

1.  **Cost**: Deploying and maintaining LLMs can be costly, involving
    expenses related to computing resources, licensing fees, and
    personnel for model management and integration.
2.  **Skill Gaps**: There might be a lack of expertise within the
    organization to effectively deploy, manage, and leverage LLMs,
    requiring significant training or hiring of specialized personnel.
3.  **Change Management**: Integrating LLMs into workflows can require
    significant changes in processes and possibly the organizational
    culture, leading to resistance among staff.

### Ethical and Legal Challenges

1.  **Accountability**: Determining accountability for decisions made or
    assisted by LLMs can be difficult, raising questions about liability
    and ethical responsibility.
2.  **Compliance**: Ensuring that the use of LLMs complies with relevant
    laws and regulations, including those related to data protection
    (e.g., GDPR), is essential.
3.  **Transparency and Explainability**: LLMs are often seen as "black
    boxes," making it challenging to understand how they arrive at
    certain conclusions or decisions, which can be problematic for
    regulatory compliance and trust.

### Mitigation

Organizations can address these challenges through a combination of
technical solutions, policy development, and training. This includes
implementing robust data governance policies, investing in secure and
scalable infrastructure, continuously monitoring for bias and errors,
ensuring compliance with legal standards, and fostering a culture of
ethical AI use.

Integrating LLMs into organisational workflows holds immense potential
for efficiency gains and innovation, but navigating these challenges is
crucial for successful and responsible implementation.

![](https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b5e8e2a95bfe){width="1"
height="1"}


Authors: [Hui Yin](https://www.linkedin.com/in/hui-yin-784538213/),
[Amir Aryani](https://www.linkedin.com/in/amiraryani/)

> RAG is an AI framework enhancing pre-trained language models with
> up-to-date, reliable external knowledge.

As we discussed in our previous article "[A Brief Introduction to
Retrieval Augmented Generation
(RAG)](https://medium.com/@amiraryani/a-brief-introduction-to-retrieval-augmented-generation-rag-4bd6e50da532)",
RAG is an artificial intelligence framework that incorporates the latest
reliable external knowledge and aims to improve the quality of responses
generated by pre-trained language models (PLM). Initially, it was
designed to improve the performance of knowledge-intensive NLP tasks
([Lewis et al., 2020](https://arxiv.org/abs/2005.11401)). As more and
more people notice the advantages of RAG, it has also been applied to a
broader range of scenarios. We conducted a rapid review of RAG-related
publications and found there are mainly five categories of use cases:
Medical, Knowledge Graph, Code Comment, Chatbot, Review, and Image
processing.

<figure>
<img
src="https://cdn-images-1.medium.com/max/1024/1*HGpI_s3nvDIyHRE2_yE5kQ.png" />
<figcaption>Created by Dr Hui Yin, Jan 2024. CC-BY License
for reuse.</figcaption>
</figure>

### **Chatbot**

Chatbots have gained widespread recognition and use by the public. It
acts as an intermediary, translating user queries or requests written in
natural language and communicating with the underlying LLM to generate
responses.

The researchers incorporated RAG into the following tasks to improve the
performance of the chatbot, allowing it to provide more accurate
responses.

Question Answering ([Mao et al.,
2020](https://arxiv.org/abs/2009.08553), [Ju et al.,
2022](https://doi.org/10.48550/arXiv.2210.02933), [Shamane et al.,
2023](https://doi.org/10.1162/tacl_a_00530)) requires access to external
knowledge bases, documents, or structured data to provide accurate and
informative responses.

Text summarization ([Cai et al.,
2022](https://doi.org/10.1145/3477495.3532682), [Hofstätter et al.,
2023](https://doi.org/10.1145/3539618.3591687)), which requires an
understanding of the content and context with the assistance of external
knowledge to generate a concise and informative summary of longer text.

In certain scenarios, dialogue systems can also benefit from RAG
technology, allowing them to handle specific tasks more efficiently
([Shuster et al., 2021](https://doi.org/10.48550/arXiv.2104.07567),
[Thule et al., 2021](https://doi.org/10.48550/arXiv.2102.04643)).

### **Code Comment and Summarization**

Code comments are crucial as they provide explanatory information about
the purpose, functionality, and usage of the code, clear and informative
comments help improve code understanding.

[Yu et al., 2022](https://doi.org/10.48550/arXiv.2206.13325) proposed
BashExplainer which aims to automatically generate code comments for
Bash scripts to enhance code comprehension, especially for developers
who may struggle with understanding Bash code due to its unique
characteristics.

[Liu et al., 2021](https://doi.org/10.48550/arXiv.2006.05405) introduced
a novel retrieval-augmented mechanism for code summarization, aiming to
bridge the language gap between source code and natural language
summaries.

### **Image Processing**

Image processing covers three topics, text-to-image, image-to-text and
image synthesis.

Text-to-image is the generation of visual content (images) from textual
descriptions.

[Chen et al., 2022](https://doi.org/10.48550/arXiv.2209.14491) designed
Re-Image to address challenges in generating images for uncommon
entities. By leveraging a multi-modal knowledge base and a retrieval
step, the model significantly improves fidelity, particularly for rare
entities.

Image-to-text is the reverse of text-to-image, it converts visual
information(images) into textual descriptions, such as generating a
caption or description.

[Sarto et al., 2022](https://doi.org/10.48550/arXiv.2207.13162) proposed
a Retrieval-Augmented Transformer for Image Captioning, which combines a
knowledge retriever, differentiable encoder, and kNN-augmented attention
layer, and enhances caption quality by using explicit external memory.

Image synthesis is very different from the above two items. It creates
new images or modifies existing images, such as generating realistic
images of non-existent objects or scenes, often using generative models
such as generative adversarial networks (GAN). You can notice that some
medium posts contain very stunning images at the top of the post page
and are labelled "Generated by AI". So far, we know of only one work
that uses RAG for image captioning. Blattmann et al. proposed a
Retrieval-Augmented Diffusion Model as a semi-parametric approach for
generative image synthesis.

### **Review**

RAG enhances review generation by incorporating external knowledge, such
as product specifications and customer feedback. The retrieval mechanism
ensures that generated reviews are more informative, contextually
relevant, and aligned with the product or service characteristics,
resulting in nuanced and helpful content compared to traditional
generative models. Several works were found in this research topic, such
as [Kim et al., 2020](https://aclanthology.org/2020.coling-main.207).

### **Medical**

RAG technology has been widely used in Medical area, such as the
Summarisation of electronic health records (EHRs) ([Saba et al.,
2024](https://doi.org/10.48550/arXiv.2401.01469), [Thompson et al.,
2023](https://doi.org/10.48550/arXiv.2312.06457)), Evidence-based
medicine (EBM) ([Vaid et al.,
2024](https://doi.org/10.48550/arXiv.2401.02851)). In fact, it is
obvious that the data processed by the models in these works is text
data related to medical content, which is still an NLP task in nature,
and the external data used in this example specifically belongs to the
medical field.

### **Knowledge Graph**

Integration of Knowledge Graph in RAG is mostly done to improve the
accuracy and performance of the RAG Pipelines. This is mainly a
methodology rather than a use case. However, given the substantial
community support, it has transformed to be an important characteristic
of RAG pipelines.

The knowledge graph is a structured representation of knowledge that
captures entities, their attributes, and their relationships in a
graph-like format. It is structured data that is easy to retrieve. In
Retrieval-Augmented Generation (RAG), leveraging information from
knowledge graphs like Wikipedia can assist in generating more
contextually relevant and accurate natural language text, especially for
open-domain tasks. In addition, the RAG structure uses other external
resources depending on the specific task, such as web scraping,
retrieval of databases, existing text corpora, and external APIS. Below
are some survey papers on RAG with in-depth discussions on external
sources and knowledge graphs, [Wei et al.,
2021](https://doi.org/10.48550/arXiv.2110.08455), [Agrawal et al.,
2023](https://doi.org/10.48550/arXiv.2311.07914), [Pan et
al., 2023](https://doi.org/10.48550/arXiv.2306.08302).

### **Acknowledgement**

This research was supported by the Australian Government through the
Australian Research Council's Industrial Transformation Training Centre
for Information Resilience (CIRES) project number IC200100022.

### References

Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N.,
Küttler, H., Lewis, M., Yih, W.-t., Rocktäschel, T., Riedel, S., &
Kiela, D. (2020b). Retrieval-Augmented Generation for
Knowledge-Intensive NLP Tasks. Neural Information Processing Systems 33
(NeurIPS 2020). <https://arxiv.org/abs/2005.11401>

Mao, Y., He, P., Liu, X., Shen, Y., Gao, J., Han, J., & Chen, W. (2020).
Generation-Augmented Retrieval for Open-domain Question Answering.
*CoRR*, abs/2009.08553. <https://arxiv.org/abs/2009.08553>

Ju, M., Yu, W., Zhao, T., Zhang, C., & Ye, Y. (2022). Grape: Knowledge
Graph Enhanced Passage Reader for Open-domain Question
Answering.<https://doi.org/10.48550/arXiv.2210.02933>

Shamane Siriwardhana, Rivindu Weerasekera, Elliott Wen, Tharindu
Kaluarachchi, Rajib Rana, Suranga Nanayakkara; Improving the Domain
Adaptation of Retrieval Augmented Generation (RAG) Models for Open
Domain Question Answering. *Transactions of the Association for
Computational Linguistics* 2023.11
1--17.<https://doi.org/10.1162/tacl_a_00530>

Cai, D., Wang, Y., Liu, L., & Shi, S. (2022). Recent advances in
retrieval-augmented text generation. In \*Proceedings of the 45th
International ACM SIGIR Conference on Research and Development in
Information Retrieval\* (pp. 3417--3419)
[https://doi.org/10.1145/3477495.3532682](https://doi.org/10.1145/3477495.3532682%5D(https://doi.org/10.1145/3477495.3532682))

Hofstätter, S., Chen, J., Raman, K., & Zamani, H. (2023). Fid-light:
Efficient and effective retrieval-augmented text generation. In
\*Proceedings of the 46th International ACM SIGIR Conference on Research
and Development in Information Retrieval\* (pp. 1437--1447)
[https://doi.org/10.1145/3539618.3591687](https://doi.org/10.1145/3539618.3591687%5D(https://doi.org/10.1145/3539618.3591687))

Shuster, K., Poff, S., Chen, M., Kiela, D., & Weston, J. (2021).
Retrieval augmentation reduces hallucination in conversation. \*arXiv
preprint arXiv:2104.07567\*. <https://doi.org/10.48550/arXiv.2104.07567>

Thulke, D., Daheim, N., Dugast, C., & Ney, H. (2021). Efficient
Retrieval Augmented Generation from Unstructured Knowledge for
Task-Oriented Dialog. \*arXiv preprint\* arXiv:2102.04643.
[https://doi.org/10.48550/arXiv.2102.04643](https://doi.org/10.48550/arXiv.2102.04643%5D(https://doi.org/10.48550/arXiv.2102.04643))

Yu, C., Yang, G., Chen, X., Liu, K., & Zhou, Y. (2022). BashExplainer:
Retrieval-Augmented Bash Code Comment Generation based on Fine-tuned
CodeBERT. <https://doi.org/10.48550/arXiv.2206.13325>

Liu, S., Chen, Y., Xie, X., Siow, J., & Liu, Y. (2021).
Retrieval-Augmented Generation for Code Summarization via Hybrid GNN.
<https://doi.org/10.48550/arXiv.2006.05405>

Chen, W., Hu, H., Saharia, C., & Cohen, W. W. (2022). Re-Imagen:
Retrieval-Augmented Text-to-Image
Generator.<https://doi.org/10.48550/arXiv.2209.14491>

Sarto, S., Cornia, M., Baraldi, L., & Cucchiara, R. (2022).
Retrieval-Augmented Transformer for Image
Captioning.<https://doi.org/10.48550/arXiv.2207.13162>

Blattmann, A., Rombach, R., Oktay, K., Müller, J., & Ommer, B. (2022).
Retrieval-augmented diffusion models. \*Advances in Neural Information
Processing Systems, 35\*, 15309--15324.

Kim, J., Choi, S., Amplayo, R. K., & Hwang, S. (2020).
Retrieval-augmented controllable review generation. In Proceedings of
the 28th International Conference on Computational Linguistics (pp.
2284--2295).<https://aclanthology.org/2020.coling-main.207>

Saba, W., Wendelken, S., & Shanahan, J. (2024). Question-Answering Based
Summarization of Electronic Health Records using Retrieval Augmented
Generation.<https://doi.org/10.48550/arXiv.2401.01469>

Vaid, A., Lampert, J., Lee, J., Sawant, A., Apakama, D., Sakhuja,
A., ... Nadkarni, G. (2024). Generative Large Language Models are
autonomous practitioners of evidence-based medicine.
<https://doi.org/10.48550/arXiv.2401.02851>

Thompson, W. E., Vidmar, D. M., De Freitas, J. K., Pfeifer, J. M.,
Fornwalt, B. K., Chen, R., ... Miotto, R. (2023). Large Language Models
with Retrieval-Augmented Generation for Zero-Shot Disease
Phenotyping.<https://doi.org/10.48550/arXiv.2312.06457>

Wei, X., Wang, S., Zhang, D., Bhatia, P., & Arnold, A. (2021). Knowledge
Enhanced Pretrained Language Models: A Comprehensive
Survey.<https://doi.org/10.48550/arXiv.2110.08455>

Agrawal, G., Kumarage, T., Alghami, Z., & Liu, H. (2023). Can Knowledge
Graphs Reduce Hallucinations in LLMs?: A Survey.
<https://doi.org/10.48550/arXiv.2311.07914>

Pan, S., Luo, L., Wang, Y., Chen, C., Wang, J., & Wu, X. (2023).
Unifying Large Language Models and Knowledge Graphs: A
Roadmap.<https://doi.org/10.48550/arXiv.2306.08302>

![](https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=111ad032653d){width="1"
height="1"}
