Photo: Tada Images/Shutterstock.com
In its first-ever developer conference in San Francisco, OpenAI announced many products to expand further and integrate generative AI into people’s daily lives.
Two main products launched at the conference are the GPT-4 Turbo and GPTs. The former is an extended and more fine-tuned version of GPT-4, while the latter is a customised agent catering to the specific needs of the customers.
GPT-4 Turbo: Next-gen GPT-4
The company introduced GPT-4 Turbo, an enhanced iteration of its text-generating AI model, GPT-4. This new version is touted to be both more powerful and cost-effective.
GPT-4 Turbo offers two variations: one dedicated to text analysis and another capable of comprehending both text and image context. The text-analysing model is now available for preview via an API, with general availability expected in the coming weeks. OpenAI has priced these models at $0.01 per 1,000 input tokens and $0.03 per 1,000 output tokens. The pricing for image-processing GPT-4 Turbo depends on the image size, with a specific example costing $0.00765.
In a blog post, OpenAI emphasised that GPT-4 Turbo is more cost-effective, offering a 3x reduction in input token prices and a 2x reduction in output token prices compared to GPT-4. The new model benefits from a knowledge base up to April 2023, a significant improvement over GPT-4, with a knowledge cutoff date of September 2021.
Moreover, GPT-4 Turbo boasts an expanded context window, offering a 128,000-token capacity, four times larger than GPT-4. This ensures more accurate and contextually relevant responses in longer conversations.
OpenAI has also introduced a “JSON mode” and additional parameters for developers, making the model more versatile and capable of returning consistent completions.
Furthermore, GPT-4 Turbo also can accept images with the help of Chat Completions API, enabling specific use cases like generating captions and analysing and detailing real-world objects.
While the company is working on GPT-4 Turbo, it has not forgotten GPT-4. The company has launched an experimental access program for fine-tuning GPT-4.
The company is also releasing a new version of GPT-3.5 Turbo, supporting a 16K context window. This platform, according to OpenAI, will show a 38% performance boost for completing tasks such as generating JSON, XML and YAML.
GPTs: Custom AI agents
OpenAI also introduced custom versions of ChatGPT for specific case uses without the need for coding.
OpenAI’s new custom agents, GPTs, will soon be available through the GPT Store. While specific details about the store’s operation are currently limited, OpenAI has pledged to compensate creators based on the usage of their GPTs. These customised GPTs will be accessible to ChatGPT Plus subscribers and OpenAI enterprise customers for internal use.
During the presentation, OpenAI showcased a specific use case of GPT when a bot called “Creative Writing Coach” analysed a PDF. Yet another bot named “Event Navigator” analysed a PDF about the event and answered the questions.
Individual GPTs can be connected to the internet, DALL-E and the Code Interpreter tool of OpenAI. Customers can also upload their specific data in the “Knowledge” section. Furthermore, to integrate external third-party data like email or e-commerce orders, OpenAI has another tool called “Actions”.
Enterprise users can access GPTs from Wednesday.
Other products
The DevDay saw the launch of a few other products:
- Assistants API: OpenAI has unveiled Assistants API, specialised AI entities capable of executing specific tasks, leveraging extra knowledge, and utilising models and tools. The API features Code Interpreter, Retrieval, and function calling to simplify complex tasks, enabling developers to build high-quality AI applications.
- Custom models: The company is also launching the Custom Models program, offering organisations more customisations for AI needs. This initiative caters to domains with large proprietary datasets, enabling dedicated collaboration with OpenAI researchers. The program allows modification at every step of the model training process, including domain-specific pre-training and custom reinforcement learning post-training.
Apart from that, OpenAI allowed developers to integrate DALL·E 3 directly into products or apps. The company recently integrated DALL·E 3 into ChatGPT Plus and Enterprise.
OpenAI also launched Copyright Shield to protect consumers against copyright claims. With this, OpenAI will defend customers and will even pay for the legal costs incurred in cases involving copyright infringements.
In the News: Minecraft emerges as the go-to target game for hacker groups
