Generative AI is exciting because it provides what users want: answers, not documents. And more than that, expertise. Large language models like ChatGPT show expertise in many tasks, even simple programming ones. There are limitations, ChatGPT only knows information before September 2021. ChatGPT thinks the current “state of the art” in search consists of Pagerank, Semantic Search, BERT, and transformers.
ChatGPT’s “knowledge” is mostly from public sources, like the web. To get responses that consider your business, you need to provide all the relevant content and context, which is hard and time-consuming. More importantly, ChatGPT lacks the tone and language of your business. Profit, leadership, product, vacation, and success is different for every business.
The opportunity is integrating your own data with ChatGPT and using the capabilities of conversational AI to unlock a wide variety of use cases — from question-answering to insight extraction and workflow automation.
For example, we begin prompting ChatGPT-3.5 via openai.com “Tell me about Sid Probstein”.
As of my last update in September 2021, Sid Probstein is not a widely-recognized public figure in fields like academia, entertainment, politics, or technology. It is possible that he is a private individual or a professional in a specific industry without wide public recognition. It is also possible that he gained prominence after my last training data was updated.
Biographies are one of the few areas where Google search is still far better than ChatGPT. But here is a summary generated with the same prompt using the Swirl Galaxy interface and Swirl RAG to select and aggregate the top relevant web results with secured OneDrive results.
Based on the provided background information, Sid Probstein is a guest on the Vector Podcast hosted by Dmitry Kan. They discussed various topics, including the challenges of enterprise search in the post-cloud era, the complexities of cross-silo search due to permissions across different data silos, the zero-code configuration of connectors in Swirl using JSON path and developer API documentation, and the contextual re-ranking of search results using large language models. Towards the end of the conversation, Dmitry mentioned using a service called Clearword to transcribe recordings. Dmitry asked how quickly the transcript could be indexed and searched using Swirl. Sid responded that the transcript was already indexed. Clearword emailed the transcript to both shortly after the recording ended, and Microsoft Outlook indexed it within seconds of arriving in Sid’s inbox. Sid verified this by copying some text from the transcript and pasting it into Swirl, which returned the link to the email message containing the transcript and the searched phrase. Sid also mentioned that the concept of metasearch, where Swirl sends queries to existing search APIs (application programming interfaces) and re-ranks the results from everything, is a game-changing approach. It eliminates the need for creating additional repositories and saves users a significant amount of time without requiring a major IT project. If interested in exploring Swirl, Sid provided instructions to get started with just two commands via Docker on their GitHub page: https://github.com/swirlai/swirl-search.
Imagine dynamically searching and retrieving information from external and internal sources like HubSpot deals, Atlassian knowledge base articles, Slack, ServiceNow documents and getting answers.
Please contact email@example.com to learn more.