Whiteboard Technical Series: Large Language Models

0:09 generative AI is revolutionizing and

0:12 humanizing AI Ops interfaces and

0:14 Associated AI native networking

0:17 platforms today in the tech whiteboard

0:19 series we talk about generative Ai and

0:22 large language models also known as llms

0:25 we'll also cover a technique called

0:26 retrieval augmented generation or rag

0:29 that enhances llms quality and output

0:32 llms are a language model Adept at

0:35 understanding and generating human

0:36 language by employing Deep learning

0:38 through neural networks they analyze

0:41 vast amounts of text Data accessing

0:43 billions of parameters to discern

0:45 patterns understand structure and

0:47 predict words or generate sentences

0:50 chatbots Incorporated AI techniques like

0:52 natural language processing or NLP and

0:55 transfer learning which accelerated

0:56 their evolution to better understand

0:58 users natural language input previous

1:01 videos in this series have covered NLP

1:04 now the recent advances in llms enable

1:06 the retention of Greater context and

1:08 complexity in human to agent

1:10 conversations expanding their utility

1:12 and Integrations llms have some

1:15 limitations such as being confined to

1:17 their training data up to their last

1:19 update or generating erroneous

1:21 predictions called

1:23 hallucinations techniques like prompt

1:25 engineering and rag are used to mitigate

1:27 these issues and enhance response

1:29 accuracy and freshness the input given

1:32 to an llm is called a prompt prompt

1:35 engineering is the art of formulating

1:37 precise questions to get better safer or

1:39 more accurate responses prompts can be

1:42 refined enhanced or sanitized before

1:45 being submitted new information can be

1:48 integrated from external sources and

1:49 databases using rag this approach

1:52 ensures responses remain relevant

1:54 accurate fresh and trustworthy enhancing

1:57 the llm output with rag is a multi-stage

2:00 process of indexing fresh retrieval and

2:03 generation first all relevant documents

2:06 such as Juniper Network's public

2:08 documentation must be chunked and

2:10 indexed into a vector database next we

2:13 retrieve relevant documents based on the

2:15 user's input query finally the relevant

2:18 documents are combined with the prompt

2:20 as additional context the combined text

2:22 and prompt are then passed to the llm to

2:24 generate the accurate and contextually

2:26 relevant response the missed AI engine

2:30 features Marvis a conversational Network

2:32 assistant enhanced with an llm and

2:35 Juniper Network's public documentation

2:37 via rag raising its understanding and

2:39 capabilities hence Marvis can now answer

2:42 custom inquiries regarding Juniper

2:44 products and services going Beyond its

2:47 troubleshooting capabilities for your

2:48 Juniper missed manag deployments teams

2:51 collaborate better with AI tools that

2:53 have enhanced understanding of their

2:55 intent and desired outcomes Juniper

2:57 Networks is pioneering AI native Network

2:59 working with tools like NLP llms and rag

3:03 democratizing access and demistifying

3:05 complex operations users and operators

3:08 at all levels can work at their own pace

3:11 while rapidly learning understanding and

3:13 driving change