At 10:00 am Pacific Standard Time on November 6th (2:00 am Beijing Time on November 7th), OpenAI will host the first global developer conference "OpenAI DevDay" in San Francisco.
OpenAI announced on its official website that the one day event will invite hundreds of developers from around the world to preview new tools and exchange ideas with the OpenAI team. OpenAI CEO Sam Altman revealed on social media that he will not release GPT-5 or GPT-4.5 or similar big models, but "will launch some great new things".
For developers, what will be new at this conference?
Based on comprehensive media reports and online revelations, we have compiled a conference outlook:
Cheaper AI model: or up to 20 times cost reduction for developers
According to Reuters, OpenAI is expected to launch a major upgrade for developers in November, allowing developers to develop ChatGPT based applications at lower costs and faster speeds, including adding memory storage APIs to developer tools, which can significantly reduce API costs by 95%. Overall, Reuters stated that this upgrade can theoretically reduce development costs by up to 20 times.
AI model has more powerful functions: multi modal or visual tools will be launched
According to Reuters, OpenAI may launch visual tools in November, which can process and generate different types of media besides text, such as images, audio, and video. Developers can use this to build applications with the ability to analyze and describe images, and apply them to fields such as healthcare, finance, e-commerce, and entertainment.
OpenAI is continuing to make efforts in the field of multimodal AI. In September, OpenAI launched multimodal features such as voice dialogue and image recognition on ChatGPT, and may launch a multimodal large model GPT-vision in the future.
Or support customized AI assistants to open critical moments in the AI consumer market
In August, OpenAI announced on its official website that it would open the GPT-3.5 Turbo fine-tuning function, allowing developers to customize ChatGPT according to their own needs and scenarios. It also stated that it would open the GPT-4 fine-tuning function in the autumn. The fine-tuned model performs better in specific scenarios, and according to OpenAI testing, the GPT-3.5 Turbo fine-tuned version has outperformed GPT-4 in some tasks.
More importantly, if the GPT-4 fine-tuning function is opened, it is convenient to customize large models. Users can customize chat robots and generate personalized AI assistants, allowing everyone to have their own "Jarvis". User CHOI posted online stating that OpenAI will release a Gizmo tool specifically designed for creating, managing, and selecting custom chat robots.
Chen Baojian of Open Source Securities predicts that with the release of GPT-4 fine-tuning function in the future, the performance of the general large model in vertical scenarios may be further optimized.
If this revelation is true, it will become a critical moment in the AI consumer market, as OpenAI is developing into a mature UGC platform where users can create and share any AI intelligent agent, according to Jim Fan, a research scientist at NVIDIA. OpenAI is also a super collection of RPA, custom AI, and plugin stores.
Other explosives
According to internet users, the company will also launch new enterprise subscription services and workspace management functions.
Among them, the team's planned price is $30 per month, and the annual package is $25 per month (a total of $300 per year), but it must include at least three users. The team plans to provide additional benefits such as unrestricted fast GPT-4 access, 4x long context, and unlimited use of advanced data analysis models.
The workspace management function provides the separation of individuals and workspaces, as well as the ability to define roles, departments, or other metadata. It should be able to integrate external ERP and CRM systems.
In addition, it is said that OpenAI will also provide a "connector" that allows users to connect third-party cloud drives (including Google Drive and Microsoft 365) to ChatGPT, which allows documents, spreadsheets, or presentations to be attached to chats or used as conversation context.