By Sarah Luger | Apr 19, 2023
Editor’s note: This story was first published in March 2023 in our internal newsletter and is being released for all readers as part of our AI April program. Learn more about our two AI April exploratory live panels here (opens in a new window)!
Key insight:
The generative AI product landscape continues to evolve with the arrival of the OpenAI Plugin store and Databricks LLM. The OpenAI Plugin store positions ChatGPT as an alternative default internet interface, allowing users to access information without conventional online engagement. Databricks’ Dolly LLM is a step toward making language models widely accessible. The structure of the generative AI product ecosystem continues to rapidly evolve.
The last two months have contained more AI announcements than a typical year and will likely continue to impress. Significant stories this week include the release of OpenAI’s Plugin toolkit that supports numerous useful ChatGPT applications, and the release of Databricks’ smaller and yet competitive open source LLM.
The OpenAI Plugin store announcement has numerous ramifications. This move positions ChatGPT as an alternative default internet interface, allowing users to access information without conventional online engagement like searching Google or visiting web sites. The rate of adoption of this ChatGPT iteration is worth watching. If the trendline of OpenAI’s recent releases continues, this could mean the end of Google’s search dominance.
With its Plugin store, OpenAI can now avoid the gatekeeper issue that has challenged other ecosystems. Platforms such as the Apple App and Google Play stores have launched successful app development ecosystems. While the hyperscalers app stores primarily act as gatekeepers for larger ecosystem, they have also allowed developers to test and roll out products in an agile manner supporting millions of services that have both flourished and floundered.
The existing toolkit which accompanies access to the Open Al ChatGPT natural language processing (NLP) platform is a multi-pronged offering that includes a host of pre-built plugins that developers can use to add Al to their existing apps. The plugins are focused on different categories, such as message ordering and knowledge retrieval. The Al plugin toolkit is distributed under the Apache 2.0 open-source license and is currently in private beta. It has been supported by the software company Ayla Networks, which was acquired by Open Al in 2018.
Widely known for its big data and AI analytics platform, Databricks entered the LLM race this week by releasing their Dolly LLM. Databricks claims it is based on a “completely open-source six-billion-parameter model from EleutherAI” and has similar functionality to Stanford’s Alpaca. Alpaca has been used for setting up chatbots and is primarily a data gathering tool. Dolly was trained on a corpus of human-like questions and answers called Alpaca 2. It is important because one of the biggest hurdles in deploying language models into production is gathering high-quality training data, which can be difficult and expensive.
Dolly is a step toward making language models widely accessible, as it enables newcomers to get started without going through training or gathering training data. It has the potential to democratize language models and make them easier for companies to deploy or improve their products using generative models. AI hype is inevitable, but these last few weeks have contained a fair share of AI hope.
Dr Sarah Luger, Technology Group Principal, Orange Silicon Valley
Sarah believes that consumers should be able to purchase the products they want in the language of their choice. Her professional background blends NLP (Natural Language Processing) engineering and product architecture, responsible AI, conversational AI, emerging machine translation technologies for under resourced languages, and she built several “first-of-a-kind” NLP solutions including automated email and text response experiences that are mainstream today. She leads efforts at Orange Silicon Valley in NLP and Responsible AI. Dr Luger received her BS from Swarthmore College and both a Master of Science and PhD in Informatics from University of Edinburgh in Edinburgh, Scotland.