OpenAI announces improvements to its AI models and developer offerings.
The highlight is the improved GPT-4 model (gpt-4-0125-preview), which is now more efficient and less "lazy," according to OpenAI. The company is also announcing new embedding models, price reductions, and new API key management tools.
The GPT-4 update fixes a problem known as "laziness" - situations where the model fails to complete tasks. In November and December, some developers complained that ChatGPT with GPT-4 had become lazy, writing incomplete code.
With the new version, gpt-4-0125-preview, the model should now complete tasks such as code generation more thoroughly and give incomplete answers less often. OpenAI promises greater accuracy and more reliable results, without going into detail about what has been fixed.
GPT-4 Turbo with Vision will be generally available "in the coming months", and OpenAI's moderation model gets an update, now being the "most robust moderation model to date," according to OpenAI.
OpenAI cuts GPT-3.5 prices - again
OpenAI has lowered prices for the GPT-3.5 Turbo model to make it easier for customers to scale, according to OpenAI. Input prices have been reduced by 50 percent to $0.0005 per 1K token and output prices have been reduced by 25 percent to $0.0015 per 1K token. This is the third price reduction in a year.
In addition to the GPT-4 and GPT-3.5 updates, OpenAI has introduced two new embedding models: text-embedding-3-small and text-embedding-3-large.
According to OpenAI, the small model offers better performance at a much lower price, while the large model generates high-dimensional embeddings for even more accurate results.
Embedding models translate content into a sequence of numbers that represent the underlying concepts in AI models. They capture the relationships between different data points and represent them in a lower dimension.
These vector representations enable AI models to perform complex tasks such as clustering, querying, or classification. Embedding models is an important step in preprocessing data for AI applications, so higher embedding performance at lower cost is a big plus in AI development.
New API Key Management Tools
To give developers more control and oversight over API usage, OpenAI introduces new management tools. Developers can now set permissions for API keys and track usage at the key level. According to OpenAI, these new features make it much easier to manage and analyze API usage, especially in larger organizations.