Our Thoughts on OpenAI DevDay
🚀 New Era Unleashed: OpenAI DevDay's Major Shifts for Creators & Developers, Putting Thin GPT Wrappers on Thin Ice! 🌟🧊
OpenAI's inaugural developer conference in San Francisco on November 6, 2023, has sparked a wave of excitement across the AI industry. In the days following this landmark event, there's been a vibrant buzz as everyone eagerly shares their perspectives on the innovative features introduced by OpenAI. These advancements are set to bring a new era of possibilities to the foundational model and the broader LLM market, igniting imaginations and inspiring new frontiers in AI development.
We are excited to share some of our thoughts!
😲 Are there surprises for you at OpenAI DevDay?
Hanhan:
To a certain degree. I’m surprised by how suddenly OpenAI expanded the product offering from B2B inference APIs and a simple web platform to a whole suite of new tools. But it makes sense in hindsight as they process the vast majority of the AI-enabled software, and they can learn user behavior from the data patterns sent through the API.
Duanduan:
Larger context length, more control with JSON model & multiple function calling, new modality APIs, corporate level model customization service, higher rate limits with cheaper pricing, GPT store… It seems like OpenAI tries to do it all, from enterprise licensing, to customization “consulting-like” business, to API for developers, to platform for any creators. They aim not just to be a part of the AI ecosystem but to be a central hub for the AGI future.
🛠️ What common tools were you using with LLM before?
Hanhan:
On a personal level, I have tried using MLFlow for tracking and a little bit of LangChain. MLFlow’s LLM features are experimental, and you can see it is quite geared toward traditional ML. I haven’t touched LangChain in great depth, but the abstraction it provides is excessive for simple projects. It could be useful if you’re working with Vector Databases or some other tools. Its OpenAI integration was broken for a day after DevDay when OpenAI changed their endpoint and client library.
Duanduan:
I was deep in the weeds with LangChain, creating small GPT wrapper projects like wellness research suggestions, movie & show summarization, PDF organization tools. And it became very clear that all those projects are obsolete now with the capabilities of instructing the GPT to talk in a certain style with knowledge you provided. As Hanhan you mentioned about Vector Database, I wonder if companies like Pinecone (or open-sourced Weaviate) worry about OpenAI coming out with a vector database on-demand service sometime in the near future.
💖 What are your favorite new releases?
Hanhan:
The JSON mode and function calling. I think it’s a common usage. It incorporates the functionality of several tools in one API and reduces the cost in development and maintaining the LLM chains. LangChain has built a quite long prompt for this to be used in conjunction with Pydantic. OpenAI might have learned from LangChain or user feedback and prompt signals. Not everyone wants to build just a chatbot, and most programmers want to treat the LLM APIs as RPCs that return a JSON object for other business logics.
Duanduan:
In my personal experience, the seamless integration of DALL•E, file upload capabilities, web browsing, and analytical functions stands out distinctly. Although this integration was anticipated, the rapid pace at which OpenAI executes these advancements remains impressive. However, when considering OpenAI's perspective, the release of GPT models and the GPT Store holds significant importance. This strategic move allows for the aggregation of traffic from users, developers, and creators onto a single platform. In turn, this strategy yields a vital resource for OpenAI: data and insights. It's important to clarify that this does not pertain to users' or developers' proprietary data, as OpenAI has firmly stated its policy against using such data for training purposes. Rather, the focus is on market signals derived from developers who test the market with innovative tools, potentially leading to new business ventures, and from users' engagement and interactions with these tools.
⛰️ How do you think this event will impact the landscape of AI?
Hanhan:
User stickiness is crucial for a platform business. A single API endpoint with one function couldn’t provide that. The new OpenAI product features enhance stickiness by integrating functions once reliant on third-party tools directly into the same webpage or through API calls. I surmise that the end goal of OpenAI is to be the bedrock of an ecosystem like Apple to its smartphones, where the revenue from services surpasses those of Mac or iPad for years. OpenAI is trying to answer the open-source AI movement supported by HuggingFace and Meta, by:
Enabling a greater community that can contribute to, and monetize with the GPT ecosystem, without having programming knowledge;
Maintaining control over the customer channel and data, therefore could exercise significant influence over the product monetization process and product development process; and,
Potentially enforcing an industry standard (Apple has a suite of tools and two programming languages for iOS/Mac developers).
I anticipate pushback from users favoring multi-LLM provider solutions or local deployments. It remains uncertain whether the foundational model market will be dominated by a single winner or will be well-segmented, accommodating multiple players with distinct focuses.
Duanduan:
With the bar for utilizing LLM, building tools and product lowered significantly, I believe the AI building and iterating process will be much shortened, but it also brings into sharp focus the need for meticulous attention to implementation considerations such as security, data usage, and legal implications. Also the importance of proprietary data, creative capacity, and the monetization on those data those creativity will be a significant topic going forward. The global market for data monetization is expected to grow significantly, yet the current infrastructure and data capabilities is still rudimentary. Additionally, the lack of consensus on data governance and responsibility, which could impede effective monetization strategies.
What’s your view on the AI landscape shaped by OpenAI’s DevDay? Leave us a comment below!
P.S. Here is a list of discontinued AI products by dang.ai.
Very thorough and thoughtful deep dive of the Dev Day! I can very much relate to the insight on building communities and enabling less technical people to contribute to this GPT ecosystem. Fingers crossed for a very exciting and uncertain future of how AI will reshape pretty much everything we do to interact with knowledge, thinking, and communications.