OpenAI’s first-ever developer conference marked an impressive step forward for the company’s large language model (LLM) based AI chatbot, ChatGPT. The announcement was hosted by the company CEO Sam Altman himself and there was a whole list of new features for the chatbot.
That being said, here are all the announcements made at OpenAI’s DevDay conference.
GPT 4 Turbo
A few months after announcing GPT 3.5 Turbo, OpenAI has also announced an enhanced version of GPT 4, the GPT 4 Turbo. It brings a number of improvements to the table such as a much longer context window, allowing for 128,000 tokens per prompt, up from the previous 32,000 limitation. Its knowledge cut-off has also been updated to April 2023.
The pricing is 3x cheaper than the current GPT 4 at $0.01 for 1000 input tokens and $0.03 for 1000 output tokens. There is also a version that can process images at $0.00765 for processing a 1080 x 1080 pixels image.
Another major announcement was the ability to create custom GPTs for specific tasks, whether it be for productivity or fun. This requires no coding knowledge as these GPTs can simply be built with prompts and a few instructions. Enterprise customers can also build GPTs based on a company’s internal knowledge base or connect them with external databases such as emails or calendars.
The image below shows an example of a user building a GPT that tells bedtime stories with images.
GPTs created by users will also have a marketplace called GPT Store. Over here, verified builders will be able to share their custom GPTs and browse others with different categories to choose from. The best GPTs will climb the leaderboards and will also be highlighted by OpenAI. Builders will also have the option to monetize their custom chatbots. OpenAI assures that none of the chat data will be shared with the builders.
Upgraded GPT 3.5 Turbo
GPT 4 was not the only model that received improvements. The older model, GPT 3.5 Turbo, also has a host of enhancements including a 16K token context window by default and the same information cut-off date as GPT 4 Turbo (April 2023). It is also able to follow instructions better than before and provide more reliable responses.
The cheaper pricing also comes down to GPT 3.5 Turbo with $0.0o1 for input and $0.002 for output at 16,000 tokens.
The new Assistants API from OpenAI is meant to aid developers in building an assistive AI inside their apps with specific instructions, extra knowledge, and the ability to call models and tools to perform tasks. It also includes several useful functions such as Code Interpreter, Retrieval, and function calling to save time on any heavy-lifting tasks.
Here is how OpenAI explains it in its blog post: “Use cases range from a natural language-based data analysis app, a coding assistant, an AI-powered vacation planner, a voice-controlled DJ, a smart visual canvas—the list goes on.”
Dall-E 3 API
DALL-E 3 is now available through OpenAI’s Images API with in-built moderation tools, letting developers integrate the powerful image generation tool directly into their apps and products. The moderation tools will help developers protect their applications from misuse.
The company is offering different formats and quality options, with prices starting at $0.04 per image. The output regulations range from 1024×1024 to 1792×1024 resolutions.
The AI company has announced a new Custom Models program for organizations that need more customization and fine-tuning to work with much larger datasets. This will give companies a chance to work with OpenAI researchers to train custom GPT-4 for their specific domain.
“This includes modifying every step of the model training process, from doing additional domain-specific pre-training, to running a custom RL post-training process tailored for the specific domain.”
OpenAI has introduced Copyright Shield, wherein the company will intervene and cover the expenses in the event that customers encounter legal claims related to copyright infringement. This protection extends to the generally available features of ChatGPT Enterprise and the developer platform.
- Whisper V3 has been released as a successor to OpenAI’s open-source automatic speech recognition model (ASR). It boasts enhanced performance across various languages and it will be incorporated into their API in the upcoming period.
- Consistency Decoder is becoming open-source, serving as a seamless replacement for the Stable Diffusion VAE decoder. This decoder enhances the quality of all images that are compatible with Stable Diffusion 1.0+ VAE.
- Developers now have the capability to produce high-quality speech from text using the text-to-speech API. The pricing structure commences at $0.015 per 1,000 characters of input.