Beta Update August #1 - Chat Bots

Another update from the flight deck. While everyone is enjoying their well-deserved summer breaks, we've been building great new functionality into Seaplane and delving into new possible applications for them.

In-context learning

LLMs are remarkable tools set to bring about a revolution in various industries. However, they do possess their own sets of limitations. By default, they can’t address questions that fall outside the scope of their training dataset. A concrete example is GPT-3.5, which can’t answer any questions about events happening after September 2021

Although retraining the model is an option, it requires a considerable investment of time and money. Fortunately, there is a solution: in-context learning with the support of a vector database. You can read more about it in our most recent blog post.

Seaplane Embeddings

We have added support for a native embedding function. This allows users to transform text into vector representations without leaving the Seaplane ecosystem. Removing the need for more expensive commercial embedding APIs such as OpenAI. Documentation is coming soon.

Langchain Integration

We listened to our users and the community. Everyone loves Langchain, and so do we. But while Langchain is amazing for running local experiments on your laptop, it has proven a bit harder to take those same workloads to production to serve customers

We are starting to address this concern by integrating Langchain with Seaplane. Using the great functionality of Langchain and the scalability of Seaplane.

Specifically, we added support for the seaplane embeddings, vector store, and hosted large language models to be used in Langchain functions.

For example, you can now create Conversational retrieval chains powered by the Seaplane vector store MPT-30B and our embeddings functions. The langchain integration is now available for select beta testers. Documentation is coming soon.

Chatbot tutorial

For the experienced user, you might have had an inkling based on earlier sections in this update, but you can now run LLM-powered chatbots entirely on the Seaplane platform. Combining our vector store, hosted LLM, embedding functionality, and Langchain integration to turn any data source into a chat GPT-style API endpoint.

Following this tutorial, you can learn more about it and deploy your own chatbot API. In it, we first implement a simple chatbot with in-context learning based on any text input provided through the API. In the second part, we extend the chatbot to train it on a Notion repository. Turning any Notion page into a fully-fledged LLM chat API.

Multipart form data support for POST requests

We have added support for multipart form data on POST requests. Enabling users to send entire files, such as PDFs, directly to POST-enabled applications running on Seaplane. Documentation is coming soon.

New podcast released

We invited George Williams to our second edition of the scale-up AI podcast.

In this episode, we'll discuss the evolution of technology and its impact on specialization within the industry.

George shares his insights on how mathematics has become a driving force behind AI and machine learning advancements. Additionally, we'll explore the emergence of different roles within the tech landscape and how they contribute to the growth of AI.

Join us for a fascinating conversation as we uncover the wonders of artificial intelligence with George Williams. Listen on Spotify or wherever you get your podcasts!

Thanks for reading and keep an eye out for our next update in two weeks with some exciting new developments!

‍