Due to the highly unstructured nature of Large Language Models (LLMs), there are thought and market shifts taking place on how to implement LLMs. Why do I say LLMs are unstructured? LLMs are to a large extent an extension of Conversational AI. Due to the unstructured nature of human language, the input to LLMs are conversational and unstructured, in the form of Prompt Engineering. And the output of LLMs is also conversational and unstructured; a highly succinct form of natural language generation (NLG). LLMs introduced functionality to fine-tune and create custom models. And an initial approach to customising LLMs was creating custom models via fine-tuning; of which there is a resurgence with OpenAI's no-code fine-tuning dashboard. However, innovation around RAG and Prompt Engineering has taken off with many supporting technologies being available. #LargeLanguageModels #PromptEngineering #ConversationalAI medium.com/@cobusgreyling/em…

How can we evaluate, measure and quantify the performance of a RAG pipeline? Retrieval Augmented Generation (RAG) is a very popular framework or class of LLM Application. The basic principle of RAG is to leverage external data sources to give LLMs contextual reference. In the recent past, I wrote much on different RAG approaches and pipelines. But how can we evaluate, measure and quantify the performance of a RAG pipeline? Well, considering the image below, the answer seems to be a data-centric approach, where generation and retrieval quality is measured separately. With Faithfulness and Relevance constituting Generation, and Precision and Recall in the case of Retrieval. Link to the full article and code example: cobusgreyling.medium.com/rag… #LargeLanguageModels #PromptEngineering #RAG

It does seem that the now discontinued OpenAI Classifier for Identifying "AI-Written Text" was a premature effort to illustrate responsibility and self-regulation. In the updated article below I show how inaccurate the API was, and how, at the time, the results from the ChatGPT model was significantly better. Even-though detecting AI written text was not a focus area of ChatGPT. I also argue the near impossible task of identifying LLM generated text, especially considering to what extent LLMs can be instructed to respond in a certain way. #LargeLanguageModels #GenerativeAI #PromptEngineering cobusgreyling.medium.com/ope…

The @vercel Playground has a chat and prompt interface. But what I like the most is the fact that the Vercel Playground acts as a LLM aggregator...you can have instant access to a multitude of models. @jaredpalmer sdk.vercel.ai/prompt

An Overview of Google Vertex AI With Generative AI Studio Google Vertex AI recently added a Generative AI Studio. The aim of Google Vertex AI is to allow users to build, deploy & scale machine learning (ML) models faster. Together with fully managed ML tools for any use case. Read more here: cobusgreyling.medium.com/an-…