GenAI and LLMs show no sign of stopping since last year. Whereas before most people thought it didn't matter much or were scared for the future. Now everyone seems to see the potential of LLMs. Last year we explored whether or not ChatGPT was coming for developer Jobs. We found it wasn't, and there are many areas where developers are critical for pushing ChatGPT to deliver complex solutions
Today, we reflect on the progress made since the last Drupal Dev Days.
We explore the explosion of models both closed and open source and how larger context windows impact the predictions from last year. We will evaluate if LLMs are any better at producing Views or modules from natural language prompts compared to last year. We will look at the state of "Naïve RAG" and improvements to the RAG framework. We will also delve into the importance of validating prompts, chains, and AI-driven processes through thorough reporting and human feedback
I will provide a surface-level demo of the modules in action, followed by a deep dive into how they work to help the audience understand the potential applications of using Drupal. Lastly, we will explore where this can go in the future.
Tools explored (to be updated):
- ChatGPT and LLMs
- Vector search through Pinecone
- Embedding
- Concept behind Langchain
- Opensource alternatives
- https://www.drupal.org/project/search_api_ai
- https://gitlab.com/freelygive/views_chatgpt
- https://www.drupal.org/project/openai
- https://www.drupal.org/project/ai_interpolator