OpenAI’s first-ever developer conference, DevDay 2023, showcased its aim to shape the future of AI with a range of groundbreaking innovations that could revolutionize the tech industry.
The latest models and improved features are designed to provide more control to developers, which will lead to the emergence of a new and unique ecosystem of applications.
Key new additions and improvements highlighted in OpenAI’s inaugural DevDay include:
- GPT-4 Turbo
- New Assistants API
- New Modalities in the API
- Model Customization
- Copyright Shield
- Custom GPTs
- Lower Prices
GPT-4 Turbo
GPT-4 Turbo comes with a 128K context window that can fit more than 300 pages of text in a single prompt and will be offered at a cheaper price. Turbo provides more control, especially in the context of model inputs and outputs.
Turbo updates the function calling feature by including the ability to call multiple functions in a single message. For example, users can “send one message requesting different actions, like opening the car window and turning off the A/C.”
It also provides improved instruction following (e.g., generating specific formats) and supports JSON mode, which is “useful for developers generating JSON in the Chat Completions API outside of function calling.”
Besides this, OpenAI will launch a feature enabling the return to log probabilities, which will be “useful for building features such as autocomplete in a search experience.” It will launch a new beta feature that “enables reproducible outputs, which is useful for use cases such as replaying requests for debugging, writing more comprehensive unit tests, and generally having a higher degree of control over the model behavior.”
The company will also release an updated GPT-3.5 Turbo that supports a 16K context window by default, improved instruction following, JSON mode, and parallel function calling.
New Assistants API
This feature provides new capabilities such as Code Interpreter, Retrieval, and Function Calling. OpenAI clearly states that data and files provided to the OpenAI API are never used to train their models, and developers have the option to delete the data.
The new assistant API introduces persistent and infinitely long threads, allowing “developers to simply add each new message to an existing thread, hand off thread state management to OpenAI, and work around context window constraints.” The assistant API will enable developers to embed agents into solutions, building agent-like experiences within their own applications and providing customized solutions while sharing revenues with OpenAI.
New Modalities in the API
New API modalities include:
- Text-to-speech (TTS) which allows developers to generate human-quality speech from text.
- DALL·E 3 incorporates built-in moderation to help developers protect their applications against misuse. Companies such as Snap and Coca-Cola have used DALL·E 3 to generate images and designs for their customers and campaigns.
- GPT-4 Turbo with Vision will be able to “accept images as inputs, enabling use cases such as generating captions, analyzing real-world images in detail, and reading documents with figures.”
Model Customization
Custom models create an experimental access program for GPT-4 fine-tuning. OpenAI is launching a Custom Models program, providing an opportunity for selected organizations to work with OpenAI researchers to “train custom GPT-4 in their specific domain and provide organizations with exclusive access to their custom models.”
OpenAI also highlights that custom models will not be served or shared with other customers, or used to train other models. Moreover, the proprietary data provided to OpenAI to train such models will not be reused in another context.
Copyright Shield
OpenAI will now provide a Copyright Shield, like Microsoft and Google, which includes defending OpenAI’s customers and paying costs incurred in case of legal claims around copyright infringement. This will be applicable to the features of ChatGPT Enterprise and the OpenAI developer platform, but not to its free ChatGPT tool.
This implies that there will be an indemnification clause in the contract wherein OpenAI will defend a user against whom a copyright infringement claim (arising from the output generated by OpenAI’s tools) is made and pay for damages resulting from such claims. This may encourage more developers to choose OpenAI.
Custom GPTs
OpenAI will launch customizable GPTs to create a tailored version of ChatGPT for a specific purpose and share it with others. The GPT Store will be launched later this month and will feature creations by verified builders.
Chats with GPT will not be shared with builders, and OpenAI will provide enhanced privacy controls to users, such as choosing whether data can be sent to a third-party API if GPT uses one, providing the option to opt the user’s account out of model training, verifying builders’ identities, and preventing users from sharing harmful GPTs.
Lower Prices
OpenAI will now provide cheaper APIs and higher rate limits. Besides this, OpenAI reiterates its enterprise privacy policies to assure that it abides by the principle of purpose limitation.
Recently, OpenAI LP (their for-profit Limited Partnership) has also revealed that the company is working on GPT-5 and plans to raise more funds from Microsoft.
Conclusion
The field of AI is evolving from an imperative technology to a declarative technology where users can simply state their desired outcome, and the tool provides it.
The field began with ChatGPT and ChatGPT API, which had user constraints, and has since progressed to encompass tools like ChatPDF (designed to answer questions about a document provided by the user), ChatGPT Wrappers (for instance, Jasper), and CanvaGPT (which allows users to build Canva posters). Now, customizable GPTs will encourage developers to build innovative and improved solutions with greater efficiency.
With these announcements at DevDay, OpenAI is paving the way for AI-driven innovation and empowering the digital world with advanced tools and resources.
They are also fostering an OpenAI-centric ecosystem that is magnetizing individuals and businesses, making AI an integral part of their daily usage while adding further layers of abstraction.
References
- https://openai.com/blog/new-models-and-developer-products-announced-at-devday
- https://openai.com/blog/introducing-gpts
Ajay Mago, Managing Partner at Maxson Mago & Macaulay, LLP (EM3 Law LLP).
Disclaimer: This publication is for information purposes only and should not be construed as legal advice or a substitute for legal counsel. This information is not intended to create an attorney-client relationship. Do not send us any unsolicited confidential information unless and until a formal attorney-client relationship has been established. EM3 Law is under no duty of confidentiality to persons sending unsolicited messages, e-mails, mail, facsimiles and/or any other information by any other means to our firm or attorneys prior to the formal establishment of such relationship. The views and opinions expressed herein are those of the author(s) and do not necessarily reflect the views of the firm.