Beta Update July #2 - Lots of little improvements!

Time for another update from the Flightdeck. We have reached cruising height and are deploying AI-infused applications for customers.

The theme of this second update of July is lots of little improvements. While the last update brought you the Vector Store and Object Store. This update focus on general Developer Experience improvements.

DevEx Improvements

• Quick Start Guide Get started without all the fuzz using our quick-start guide. The internal record for deploying a hello world stands at 30 seconds. Can you beat it?
• Default Project Structure At Seaplane, we aim to make deploying your pipeline as simple as possible, but even we can’t escape some basic files and structure to guide the project.
• Seaplane Secrets Pssst, did you hear the secret? Seaplane now supports a secrets service. Enabling users to specify runtime secrets and connection parameters such as API keys and database credentials in a .env file which are loaded as environment variables during deployment.
• Smaller improvements. Pipelines now return the URL on which they are available. We simplified the interface to talk to GPT3.5. Flightdeck automatically creates your first project and API key when you sign up.

Model Comparison

We are always testing the latest and greatest large language models to enable our customers to build powerful workflows using the best tech. We recently deployed MPT-30B and compared it to GPT-3.5 so that you don’t have to. We compared it based on three categories that we see most often used by our customers.

• Text Generation
• Logical Reasoning
• Classification

You can read the full report and the questions we asked it here.

Our LLM deployments are available to all customers; if you want access to try for yourself, reply to this email or sign up for the beta directly.

In addition, we are playing with an idea that we would love your input on. March Model Madness (to be held in August or September). We deploy two models and give you an interface to compare and rate them. The winner of the week moves on to the next week, where it faces the next LLM. This allows you to test dozens of models and enables us to write a report on the best available LLM models based on user feedback. Let us know if this is something you are interested in.

Data Loss Prevention In the Age Of The Large Language Model

While LLMs unlock great potential for improving productivity, the potential issues. Especially around data security and data loss prevention should not be overlooked. We wrote a blog - available here - how you can utilize Seaplane to build a wrapper around commercial LLMs to validate and vet for potential proprietary information. We are working on an open-source project on top of Seaplane implementing this pipeline. Let us know if you are interested and want to be notified once we release it.

That is it for our update this week, but keep your eyes on the horizon for more exciting updates coming soon.

Want to try Seaplane for yourself? Sign up here and get access today!

‍