OpenAI has announced a number of updates to its API, including the introduction of function calling, updated and more steerable versions of GPT-4 and GPT-3.5-turbo, a new 16k context version of GPT-3.5-turbo, a 75% cost reduction on its state-of-the-art embeddings model, and a 25% cost reduction on input tokens for GPT-3.5-turbo.
The new function calling capability allows developers to describe functions to GPT-4-0613 (here “0613” is simply the date of the update, so June 13) and GPT-3.5-turbo, and have the model intelligently choose to output a JSON object containing arguments to call those functions. This is a new way to more reliably connect GPT's capabilities with external tools and APIs.
Newer versions of GPT-4 and GPT-3.5-turbo have been fine-tuned to be more steerable, meaning that developers can have more control over the output of the model. This is done by providing the model with additional information about the desired output, such as a specific topic or style.
The new 16k context version of GPT-3.5-turbo has a larger context window than the standard 4k version, meaning that it can take into account more information when generating text. This can lead to more accurate and coherent output.
Access to the API is Now More Affordable
OpenAI has also announced a number of cost reductions for its API. The state-of-the-art embedding model is now 75% cheaper, and input tokens for GPT-3.5-turbo are now 25% cheaper. Furthermore, OpenAI has also announced the deprecation timeline for the GPT-3.5-turbo-0301 and GPT-4-0314 (The numerical model ordering denotes the date of release by month and day) models. These models will be retired on June 1, 2023.
These updates are a significant improvement to OpenAI's API and will make it easier for developers to build applications that use GPT. The new function calling capability, updated models, and cost reductions are all welcome changes that will make GPT more accessible and affordable.
It may come as a surprise that OpenAI is making its GPT-4 API so accessible. Yesterday I reported on the company apparently telling long-time partner Microsoft not to use GPT-4 in its Bing Chat AI search chatbot.
Warning Microsoft of the Risks
According to a report by the Wall Street Journal, Microsoft was warned by OpenAI, the research organization that created GPT-4, that the model was not ready for public deployment and that it could pose ethical and social risks.
GPT-4 is an impressive and powerful language model, but it is not without its flaws and challenges. One of them is that it lacks any real understanding of the world or the facts it writes about. It can generate inaccurate, misleading, or even harmful information if it is not properly guided or verified. Another one is that it can mirror and amplify the biases and prejudices that exist in its training data, such as sexism, racism, or hate speech.
OpenAI is aware of these issues and has taken steps to ensure the responsible and ethical use of GPT-4. It has limited access to the full model and has only released a smaller version called GPT-4 Playground to selected researchers and developers. It has also implemented safeguards such as filters, human oversight, and feedback mechanisms to monitor and control the outputs of GPT-4.
Nevertheless, Microsoft, which is one of the major investors and partners of OpenAI, has reportedly bypassed these restrictions and obtained access to the full model. It has integrated GPT-4 into its Bing chat mode, which allows users to interact with the search engine in a conversational way. Microsoft claims that this feature can enhance the user experience and provide more engaging and informative responses.