OpenAI DevDay: Presentation of GPT-4 Turbo and AI Assistants
What can we learn from the OpenAI conference?
OpenAI has made considerable progress over the past year and is delighted to present its latest advances at the first ever OpenAI DevDay. The event took place in San Francisco, where OpenAI has been based since its inception. The company is dedicated to growing and delivering innovative solutions to the technology industry.
OpenAI has achieved several important milestones over the past year. It launched ChatGPT as a research preview, followed by GPT-4, the most powerful model available. It also introduced voice and visual capabilities to ChatGPT, enabling it to see, hear and speak. In addition, it launched DALL·E 3, the world’s most advanced image model, and ChatGPT Enterprise, offering enterprise-level security and confidentiality.
The OpenAI API has grown in popularity, with 2 million developers using it for a wide variety of use cases. Over 92% of Fortune 500 companies rely on OpenAI products, and there are now around 100 million weekly active users on ChatGPT.
Introducing the GPT-4 Turbo: A more powerful model
OpenAI is committed to taking developer feedback on board and improving its models in line with user needs. That’s why it’s launching GPT-4 Turbo, a new, more advanced model that addresses several key areas:
1. Context length
GPT-4 Turbo supports much longer contexts, up to 128,000 tokens, compared with 8,000 tokens for GPT-4. This enables more extensive and detailed input, improving accuracy and model understanding on longer conversations or documents.
2. More control
Developers asked for more control over model responses and outputs. OpenAI has implemented JSON loading, which ensures that the model responds with valid JSON, simplifying API calls. The model also offers an enhanced call function, enabling users to call multiple functions simultaneously and follow instructions more precisely. In addition, OpenAI has introduced reproducible output, enabling consistent model behavior by specifying a seed parameter.
3. Better World Knowledge
OpenAI recognizes the importance of incorporating knowledge from today’s world into its models. It is launching retrieval, which allows users to contribute external knowledge from documents or databases. OpenAI is also committed to updating the knowledge threshold of its models, to ensure accuracy and avoid obsolete information.
4. New Modalities
GPT-4 Turbo can now accept images as input via the API and generate captions, classifications and analyses. OpenAI has integrated DALL·E 3, GPT-4 Turbo with Vision and a new text-to-speech model into the API. These additions enable advanced image generation, improved visual understanding and natural sound generation.
OpenAI has extended fine-tuning capabilities to the 16k version of GPT-3.5. It has also introduced the Custom Models program, enabling researchers to collaborate with companies to develop custom models tailored to specific domains or use cases.
6. Higher Rate Limits
OpenAI has doubled the token-per-minute rate limit for established GPT-4 customers, offering greater flexibility and enabling developers to achieve more in less time. It also introduces Copyright Shield, covering legal claims related to copyright infringement for ChatGPT Enterprise and API customers.
Revolutionizing AI with GPTs and AI assistants
OpenAI introduces GPTs, customized versions of ChatGPT that combine instructions, extended knowledge and actions. Developers can create GPTs for specific purposes and share them publicly or privately. OpenAI will feature the best and most popular GPTs in the forthcoming GPT Store, and creators will receive a share of the revenue generated.
The Assistants API simplifies the development of AI-powered assistants in applications. It includes features such as persistent threads, retrieval, code interpreter and function call enhancement. The API enables developers to create more interactive and personalized experiences for users.
OpenAI’s vision for the future
The company co-founded by Sam Altman envisions a future where artificial intelligence agents can plan and carry out complex actions on behalf of users. It believes in gradual, iterative deployment of these agents to ensure safety and societal considerations. The aim is to endow individuals with AI superpowers that enhance their capabilities and provide unprecedented assistance.
Real-world applications of OpenAI technology
OpenAI focuses not just on numbers and statistics, but also on the impact of its technology on people’s lives. It shared a video showing how people are using AI to solve real-life problems:
- One person used ChatGPT to write a heartfelt message to his father in Tagalog, expressing his love and support in a culturally appropriate way.
- ChatGPT helped a creative person overcome doubts and gain confidence in her work, serving as a useful companion.
- One scientist used ChatGPT to generate a diagram representing circulating hemoglobin, demonstrating the model’s versatility in scientific applications.
- A vegan user called on ChatGPT to identify missing ingredients in his fridge and suggest vegan recipes.
- The code interpreter, a ChatGPT feature, enabled users to create spreadsheets and automate various tasks.
- ChatGPT provided a friendly and knowledgeable virtual assistant to a person seeking information about his date of birth.
- One student and one parent found ChatGPT invaluable for answering questions and providing explanations, enhancing their learning experience.
- ChatGPT with voice integration has changed the game for someone with physical limitations, improving accessibility and ease of use.
These examples demonstrate the diversity of applications for OpenAI technology, and underline the positive impact it has on users’ lives.
OpenAI DevDay showcased the latest advances and innovations, providing developers with the tools and resources they need to shape the future of AI-based applications. OpenAI is dedicated to continuous improvement, guided by user feedback and real-world applications of its technology.
As OpenAI continues to revolutionize the AI landscape, developers can look forward to even more exciting developments and possibilities in the near future.