OpenAI LP is rolling out a new tool that will enable users to customize ChatGPT for specific tasks without writing any code.

The tool made its debut at the company’s first annual DevDay developer conference, which took place today in San Francisco. At the event, OpenAI also introduced an enhanced version of the GPT-4 language model that underpins ChatGPT. Beyond the language model market, the company is enhancing its collection of image generation and speech generation services for developers.

User-created GPTs

Users of ChatGPT’s paid tiers are gaining the ability to create custom versions of the chatbot that OpenAI refers to as GPTs. A company’s help desk team, for example, could create a GPT that answers employees’ most common technical support questions. Online retailers can build versions of ChatGPT that provide product and shipping information to buyers.

OpenAI says creating GPTs doesn’t involve any coding. Instead, users can simply enter natural language instructions explaining how they wish to customize ChatGPT. They can optionally add in a dataset that their customized chatbot should draw on when generating prompt answers.

For customers with more advanced requirements, OpenAI will make it possible to integrate GPTs with third-party services via those services’ application programming interface. A customized version of ChatGPT can, for example, retrieve information from external databases to answer questions it’s not capable of processing out of the box. GPTs can also use APIs to perform tasks in third-party applications.

The option to create customized versions of ChatGPT will start rolling out later this week. In December, OpenAI plans to launch a platform called GPT Store that will enable users to share their customized chatbots with the public. Further down the road, the company intends to let users make money from their GPTs.

For organizations that require more customizability, OpenAI is rolling out a program called Custom Models. A company that signs up for the program can have OpenAI researchers develop a version of GPT-4 optimized specifically for its requirements. “This includes modifying every step of the model training process, from doing additional domain specific pre-training, to running a custom RL post-training process tailored for the specific domain,” OpenAI detailed in a blog post.

Each customized version of GPT-4 that is developed through the program will only be made available to the customer who commissioned it. Additionally, OpenAI pledged not to share any proprietary data it receives as part of such projects.

GPT-4 Turbo 

OpenAI is rolling out the new customization features alongside an enhanced version of GPT-4. The new edition, which is known as GPT-4 Turbo, will become available to developers through the company’s language model APIs.

GPT-4 Turbo follows user instructions more reliably than its predecessor. Additionally, it quadruples the amount of data that can be included in a single prompt. It’s now possible to input up to 300 pages’ worth of information with each request.

Under the hood, OpenAI made performance optimizations that allow ChatGPT-4 Turbo to process prompts using less hardware resources. That reduction in infrastructure requirements has enabled the company to lower its API pricing. Sending prompts to GPT-4 Turbo costs three times less than with the original GPT-4, while output pricing has been halved.

OpenAI is also adding a number of more specialized features. There’s a new Turbo version of ChatGPT-3.5, a predecessor to GPT-4 that offers a more limited feature set and costs significantly less. A new Assistants API, meanwhile, combines OpenAI’s large language models with features designed to ease the task of building applications with built-in chatbots.

Whisper and DALL-E 3 

OpenAI’s DevDay conference didn’t focus solely on its large language models. The company also introduced a new version of Whisper, its open-source speech recognition model. OpenAI says that latest release provides better support for languages other than English.

In conjunction, the company announced plans to make its newest DALL-E 3 image generation model available to developers via an API. The model was previously accessible only through the paid versions of ChatGPT. DALL-E 3, the third iteration of an image generator that OpenAI debuted in 2021, can follow user instructions much more accurately than its predecessors. 

 Photo: OpenAI