To improve AI capabilities in the workplace, OpenAI announced a series of updates to its groundbreaking GPT-3.5 Turbo and GPT-4 generative AI models.
These improvements, which include the revolutionary introduction of new function call features, improved controllability, expanded context for GPT-3.5 Turbo, and a revised pricing structure, aim to provide developers with an expanded toolbox for building sophisticated, high-performance AI applications despite the complexity modern work environments.
Applications supported by OpenAI
Developers aren’t the only ones who will benefit from the latest improvements to OpenAI’s GPT models. Chances are, you’ve used a tool that implements AI advances from OpenAI.
- Microsoft has partnered with OpenAI to provide developers with AI models and enhance popular products like Bing and Office with generative AI.
- Snapchat has launched its generative AI chatbot, My AI, which uses OpenAI GPT models. The latest update of my AI can send and interpret image snapshots!
- Salesforce released the first generative AI CRM product, Einstein GPT, based on OpenAI’s “most advanced models”.
- Morgan Stanley announced a partnership with OpenAI as one of the few wealth management firms with access to the latest GPT-4 model.
- HubSpot has developed new tools like ChatSpot.ai based on OpenAI GPT-4.
- GitHub Copilot added generative AI to its platform using the OpenAI Codex to help developers, which ultimately led to a copyright infringement lawsuit.
- Stripe has integrated OpenAI GPT technology to better understand customers and reduce fraud.
- GetResponse introduced an OpenAI GPT based email generator.
- Instacart has developed an AI chatbot to help consumers shop for groceries.
Ideally, users of these and other tools built on OpenAI technology should see improvements in generative AI performance thanks to OpenAI’s GPT-3.5 Turbo and GPT-4 updates.
Improvements to GPT-3.5 Turbo and GPT-4
The following are the latest updates announced by OpenAI for the GPT-3.5 Turbo and GPT-4 models. Updates include a new function call feature in the Chat Completions API, improved controllability, expanded context for GPT-3.5 Turbo, and lower prices.
function call capability
Based on developer feedback and feature requests, OpenAI gave developers the ability to describe features for the updated models and have the AI intelligently create a JSON object with arguments for those features. This improvement allows for a more reliable connection of the GPT functions to external tools and APIs, which supports a more structured data retrieval from the model.
New functions for calling functions enable a wide range of applications, including the following:
- Creation of chatbots that answer questions by calling external tools,
- Conversion of natural language queries into function calls, API calls or database queries,
- Extraction of structured data from the text.
- The new API parameters allow developers to describe functions to the model and ask the model to optionally call a specific function.
The introduction of the function call opens up new possibilities for developers. You can integrate the GPT models more seamlessly with other APIs or external tools.
For example, a workplace app could use this feature to convert a user’s natural language query into a function call to a CRM or ERP system, making the application more user-friendly and efficient.
While OpenAI remains alert to potential security issues related to untrusted data, it suggests developers protect their applications by only leveraging information from trusted tools and include user verification steps before taking effective action.
Developers can sign up for the waitlist to get access to GPT-4.
The new GPT-4 and GPT-3.5 Turbo models offer improved maneuverability and expanded context.
Developers could use the improved controllability to design AI applications that are more attuned to the specific needs of an organization or task, such as creating more targeted business reports or creating detailed, contextual responses in customer service chatbots.
Publishing GPT-3.5 Turbo-16k can provide four times the context length of standard GPT-3.5 Turbo and support up to 20 pages of text in a single request. This expanded contextual capacity enables the AI to understand and generate responses to much larger bodies of text.
For example, in legal or academic workplaces, where documents tend to be long, this feature could dramatically improve the model’s ability to understand and summarize large amounts of text, making information extraction more efficient. Similarly, in project management applications, it could allow AI to process and understand entire project plans at once, which would help create more insightful project analysis and forecasting.
OpenAI also announced the discontinuation of previous GPT 4 and GPT 3.5 Turbo versions, with older models remaining available until September 13th. Developers were assured of a smooth transition and were encouraged to provide feedback to refine the process.
After system efficiency improvements, OpenAI passes cost savings on to developers.
The price for using the popular embedding model text-embedding-ada-002 is reduced by 75%. In addition, there is a 25% cost reduction on input tokens for the GPT 3.5 Turbo model.
I also wanted to take a second to reiterate the 75% price drop on embeds. That’s actually pretty crazy. You used to be able to embed the entire web for about $50 million, now it’s only about $12.5 million.https://t.co/fyhQVGPEi8
— Logan.GPT (@OfficialLoganK) June 13, 2023
These price reductions, coupled with improved functionality, are intended to make it easier for developers to use and experiment with these models in their applications.
Further development of the GPT models
OpenAI seems eager to continually improve its platform based on developer feedback. With the latest enhancements to its generative AI models, OpenAI offers developers new opportunities to create innovative and improved AI applications for the workplace.
The latest API updates and GPT models give developers more opportunities to build AI applications that are better suited to addressing the complexity and specificity of tasks commonly encountered in work environments.
Featured image: iama_sing/Shutterstock