Weaviate Podcast

Hey everyone! Thank you so much for watching the 66th Weaviate Podcast with David Garnitz, the creator of VectorFlow! VectorFlow (open-sourced on GH and linked below) is a new tool for ingesting data into Vector Databases such as Weaviate! There is quite an interesting End-to-End stack emerging at the ingestion layer, from retrieving data from misc. sources such as Slack, Salesforce, GitHub, Google Drive, Notion, ... to then Chunking the Text (maybe with the use of Visual Document Layout parsers like what Unstructured is imagining), extracting Metadata potentially (say the "age" of an NBA player as in the Evaporate-Code+ research) -- then sending this data off to embedding model inference and unpacking that can of worms from inference acceleration to load balancing, and finally -- importing the vectors themselves to Weaviate! I learned so much from this conversation, I really hope you enjoy listening and please check out VectorFlow below! VectorFlow: https://github.com/dgarnitz/vectorflow Chapters 0:00 VectorFlow on GitHub! 0:52 Welcome David Garnitz! 1:17 Vector Flow, Founding Vision 2:00 Billions of Vectors in Weaviate! 4:20 End-to-end data importing 6:30 Metadata Extraction in Vector Database Flows 10:15 Vectorizing 100s of millions of billions of chunks 15:58 Fine-Tuning Embedding Models 23:50 Zero-Shot Models in Metadata and Chunking 36:36 Vector + SQL 42:45 Self-Driving Databases 49:23 Generative Feedback Loop REST API 51:38 GPT Cache 55:55 Building VectorFlow

Hey everyone! Thank you so much for watching the Weaviate Podcast! I am SUPER excited to publish my conversation with Ofir Press! Ofir has done incredible work pioneering AliBi attention and Self-Ask prompting and I learned so much from speaking with him! As always we are more than happy to answer any questions or discuss any ideas you have about the content in the podcast! +Huge Congratulations on your Ph.D. Ofir! AliBi Attention: https://arxiv.org/abs/2108.12409 Self-Ask Prompting: https://arxiv.org/abs/2210.03350 Ofir Pres on YouTube: https://www.youtube.com/@ofirpress Chapters 0:00 Welcome Ofir Press 0:41 Large Context LLMs 12:38 Quadratic Complexity of Attention 19:12 AliBi Attention, Visual Demo! 24:53 Recency Bias in LLMs 28:57 RAG in Long Context LLM Training 36:27 Self-Ask Prompting 46:07 Chain-of-Thought and Self-Ask 50:47 Gorilla LLMs 58:42 New Directions for New Training Data

Hey everyone! Thank you so much for watching the 64th Weaviate Podcast with Shishir Patil and Tianjun Zhang, co-authors of Gorilla: Large Language Models Connected with Massive APIs! I learned so much about Gorilla from Shishir and Tianjun, from the APIBench dataset to the continually evolving APIZoo, how the models are trained with Retrieval-Aware Training, Self-Instruct Training data and how the authors think of fine-tuning LLaMA-7B models for tasks such as this, and many more! I hope you enjoy the podcast! As always I am more than happy to answer any questions or discuss any ideas you have about the content in the podcast! Please check out the paper here! https://arxiv.org/abs/2305.15334 Chapters 0:00 Welcome Shishir and Tianjun 0:25 Gorilla LLM Story 1:50 API Examples 7:40 The APIZoo 10:55 Gorilla vs. OpenAI Funcs 12:50 Retrieval-Aware Training 19:55 Mixing APIs, Gorilla for Integration 25:12 LlaMA-7B Fine-Tuning vs. GPT-4 29:08 Weaviate Gorilla 33:52 Gorilla and Baby Gorillas 35:40 Gorilla vs. HuggingFace 38:32 Structured Output Parsing 41:14 Reflexion Prompting for Debugging 44:00 Directions for the Future

Nils Reimers, AI researcher, discusses the collaboration between Weaviate and Cohere, temporal queries, metadata extraction, long document representation, and future directions for Retrieval-Augmented Generation in the Weaviate Podcast. They also explore the challenges of search analysis, fine-tuning language models, and user preferences in search.

Hey everyone! Thank you so much for watching the 62nd Weaviate Podcast with Atai Barkai! We are stepping into the meta with this one for a podcast about podcasts! Podcasts are one of the biggest opportunities of new technologies, starting with Whisper's ability to transcribe audio to text and advances with speaker diarization, .. the question to be explored is, What Vector Database and LLM applications can we build with this data?! What is the future of podcasting with these new technologies?! I had so much fun discussing all these ideas with Atai! As always we are more than happy to answer any questions or discuss any ideas you have about content discussed in the podcast! Thank you so much for watching! Chapters 0:00 Welcome Atai! 1:04 TawkitAI and PodcastGPT! 2:20 Chat with Podcast PodcastGPT - https://www.podcastgpt.ai/ Tawkit AI - https://twitter.com/tawkitapp Weaviate Podcast Search Demo! https://github.com/weaviate/weaviate-podcast-search

Hey everyone! Thank you so much for watching the 61st episode of the Weaviate Podcast! I am beyond excited to publish this one! I first met Rohit at the Cal Hacks event hosted by UC Berkeley where we had a debate about the impact of Semantic Caching! Rohit taught me a ton about the topic and I think it's going to be one of the most impactful early applications of Generative Feedback Loops! Rohit is building Portkey, a SUPER interesting LLM middleware that does things like load balancing between LLM APIs, and as discussed in the podcast there are all sorts of opportunities for this kind of space whether it be routing to tool-specific LLMs, different cost / accuracy requirements, or multiple models in the HuggingGPT sense. It was amazing chatting with Rohit, this was the best dive into LLMOps I have personally been apart of! As always we are more than happy to answer any questions or discuss any ideas you have about the content in the podcast! Check out portkey here! https://portkey.ai/blog Chapters 0:00 Introduction 0:24 Portkey, Founding Vision 2:20 LLMOps vs. MLOps 4:00 Inference Hosting Options 7:05 3 Layers of LLM Use 8:35 LLM Load Balancers 12:45 Fine-Tuning LLMs 17:08 Retrieval-Aware Tuning 21:16 Portkey Cost Savings 23:08 HuggingGPT 26:28 Semantic Caching 32:40 Frequently Asked Questions 34:00 Embeddings vs. Generative Tasks 35:30 AI Moats, GPT Wrappers 39:56 Unlocks from Cheaper LLM Inference

Hey everyone! Thank you so much for watching the 60th Weaviate podcast with Patrice Bourgougnon! Patrice is the creator of WPSolr, integrating AI search capabilities with Wordpress and Woocommerce. Patrice is one of the most active contributors to Weaviate, filing issues and poking holes in new releases! Patrice shared incredible feedback on Weaviate and how he sees the state of Vector Databases and Search! As always, we are more than happy to answer any questions or ideas you have about the content discussed in the podcast! Thanks for watching! Chapters 0:00 Introduction 0:45 Vector Databases and Wordpress 4:50 Weaviate Client Languages 10:00 Inference and Database Container Management 21:30 Business Opportunities for Search in Production 26:40 Testing Search Performance, “Something to sleep on” 30:50 Zero-Shot Model Ability 36:05 Make LLMs Stateful 43:46 Chatbots and Search Boxes 44:55 Mixing Models in Applications 47:00 BM25 vs. Vector Search in RETRO RAG

Hey everyone! Thank you so much for watching the 58th episode of the Weaviate Podcast! I am SUPER excited to welcome Andriy Muylar! Andriy is the Co-Founder of Nomic AI, a company fresh off a $17M series A raise! Nomic has created some incredible products such as Atlas and GPT4All! I was really impressed by Andriy's vision of the state and forecasted evolution of these topics! I hope you enjoy the podcast! As always, we are more than happy to answer any questions or discuss any ideas you have about the content discussed in the podcast! Integration Tutorial for Weaviate and Nomic AI Atlas! https://docs.nomic.ai/vector_database.html This example worked for me if you want to clone it with the podcast transcription dataset: https://github.com/weaviate/weaviate-podcast-search/blob/main/atlas-visualizer.py Check out Nomic AI here! https://home.nomic.ai/blog Chapters 0:00 Congrats Nomic and Weaviate Integration! 2:35 Welcome Andriy Mulyar! 3:05 Founding Story of Nomic AI 6:55 Understanding Massive Scale Text Data 10:14 Topic Modeling 16:30 Monitoring Model Training

Hey everyone! Thank you so much for watching the 57th Weaviate podcast with Charles Frye! Charles is an educator at Full Stack Deep Learning, one of the world's top courses on Deep Learning with lectures available on YouTube (link below)! This was one of the most thorough Weaviate podcasts published so far, covering all sorts of topics around the evolution of Deep Learning! Particularly we discussed the Retrieval-Augmented Generation stack with Vector Databases and Zero-Shot Large Language Models and how that compares to more conventional machine learning workflows and the MLOPs stack! I really enjoyed chatting with Charles and am more than happy to answer any questions or discuss any ideas you have about the content in the podcast! Thank you so much for listening! Check out Full Stack Deep Learning! https://fullstackdeeplearning.com/ Full Stack Deep Learning on YouTube! https://www.youtube.com/@The_Full_Stack Chapters 0:00 Welcome Charles Frye! 0:52 Charles’ journey into Deep Learning 3:00 Weights & Biases and MLOps 5:30 Retrieval-Augmented Generation Stack 8:58 Data Engines and AI Products 13:50 Fine-Tuning 16:35 Information Retrieval Techniques 20:10 RAG as Tool Use and RETRO 23:33 Gorilla and Fine-Tuned Tool Use 27:36 Text-to-SQL Tool Use 30:46 Generative Data Augmentation 33:05 LLM generated queries for embeddings 38:04 Long-Tail and Data Imbalance 41:45 LoRA LLM Fine-Tuning 44:50 Eigenvectors and Disentaglement 50:00 LLM for Each User 55:00 Embedding Visualization and ML Observability 58:40 GPU Utilization 1:05:05 Discord Q&A Bot App 1:16:10 Data Schema Design 1:21:25 Graph and Vector Databases 1:28:35 Future Directions in AI

Chapters 0:00 Weaviate 1.20!!! 0:40 Multi-Tenancy 35:36 PQ Rescoring 47:20 Re-Ranking, AutoCut, Rank Fusion 58:58 Cloud Monitoring Metrics