OpenAI recently shared several improvements to its overlay and text moderation models on its blog, as well as updates to its flagship GPT-4 turbo and GPT-3.5 turbo models. These innovations aim to offer developers more performance, efficiency, customization and security in their natural language or code-based applications.
OpenAI released the first version of GPT-4 in March 2023, making it available to all developers the following July. During its first developer conference, OpenAI DevDay, on November 6th, the company released a preview of the next generation of this model: Turbo GPT-4.
Different users complained that ChatGPT refused to answer their queries or only partially responded, asking them to complete the answers.
OpenAI posted to X on December 8:
“We heard all your comments about GPT4 getting lazier! We haven’t updated the model since November 11th and that’s definitely not intentional. Model behavior can be unpredictable and we seek to remedy this.”
OpenAI therefore published an updated GPT-4 Turbo preview model, “gpt-4-0125-preview”, on January 25th, indicating that “this model performs tasks such as code generation more completely than the previous visualization model and aims to reduce cases of “laziness” when the model does not perform a task.”
Bugs due to difficulties that the previous visualization model had when generating text in languages other than English were also fixed.
GPT-3.5 turbo model update
OpenAI also announced the release of a new GPT-3.5 Turbo model this week, with reduced pricing (query cost halved) and performance improvements. This update aims to provide greater response accuracy and fix coding issues found in the previous version, providing an improved experience for users.
Two new models ofincorporation at lower prices
A incorporation is a sequence of numbers that represents concepts in content, such as natural language or code. Integrations make it easier for machine learning models and other algorithms to understand relationships between content and perform tasks like grouping or searching. They power applications such as knowledge retrieval in ChatGPT and API Assistants, as well as many retrieval augmented generation (RAG) development tools.
OpenAI is releasing two new text embedding models: a smaller, more efficient model, embedding-3-small, and a larger, more powerful model, embedding-3-large. These models, more efficient and cheaper than their predecessor, embedding-ada-002, offer developers the possibility of choosing the size of embeddings according to the specific needs of their applications. This greater flexibility allows developers to optimize performance while controlling the costs associated with using AI models.
Better API Key Management
Developers now benefit from new API key management capabilities, allowing them to assign specific permissions and track API usage in greater detail. These improvements provide greater control and visibility over the use of AI resources, making project and budget management easier.
OpenAI says it plans to further improve developers’ ability to view API usage and manage API keys in the coming months, especially at larger enterprises.
Updated moderation model
OpenAI has introduced a new, more robust moderation model, allowing potentially harmful text to be identified with greater accuracy. According to OpenAI, this update demonstrates its commitment to the security and reliability of its products, ensuring a safe and positive user experience.