OpenAI LP has added new features to its ChatGPT chatbot that will enable users to search the web with Bing and generate images.

The Bing integration moved from beta into general availability on Tuesday. The new image generation capability, in turn, rolled out on Monday and is based on an artificial intelligence tool called DALL-E 3 that debuted last month.

ChatGPT is powered by GPT-4, OpenAI’s newest and most advanced large language model. By default, a language model can draw only on the information in its training dataset to answer user questions. As a result, users don’t have access to newer information that was created after the training dataset’s finalization.

OpenAI’s new Bing integration addresses that limitation. When ChatGPT receives a question to which there’s no suitable answer in its training dataset, it can now look up relevant information on the web using Bing. In its responses, the chatbot cites the websites from which it sources information.

The Bing integration originally rolled out to ChatGPT in May. However, OpenAI had to disable it a few days later after users discovered the chatbot can retrieve paywalled website content without permission. OpenAI turned the feature back on last month, but kept it in beta until this week.

The integration is one of several channels through which ChatGPT can access information not included in its training dataset. OpenAI also offers integrations, or plugins as it calls them, with a range of third-party applications. ChatGPT can use those applications to access restaurant information, e-commerce product listings and many other types of data. 

Immediately before the Bing integration exited OpenAI’s beta testing program on Tuesday, a second new feature entered the program. That feature makes it possible to generate images by entering natural language prompts into the ChatGPT interface. It’s powered by DALL-E 3, the latest addition to the DALL-E series of AI image generators OpenAI launched in 2021.

The company says that DALL-E 3 generates higher-quality images than its predecessors. Furthermore, the model can interpret user requests more accurately. This allows it to follow complicated drawing instructions that earlier DALL-E models would have struggled to understand.  

One contributor to DALL-E 3’s improved parsing capability is that it doesn’t process prompts directly. When a user types in a description of an image, an AI model analyzes the description and creates new prompts DALLE-3 can understand more easily. DALL-3, for its part, generates multiple versions of each image to increase the chances that the user’s requirements will be met. 

ChatGPT’s new search and image generation features are rolling out at a time when adoption of the chatbot is said to be surging. Last week, The Information reported that OpenAI’s annualized revenue run-rate has grown 30% since August to $1.3 billion. The company generated just $28 million in revenue during 2022. 

OpenAI’s sales growth is said to be driven primarily by purchases of ChatGPT subscriptions. For consumers, it offers a plan called ChatGPT Plus that offers benefits such as faster chatbot response times. In August, the company introduced a second subscription offering called ChatGPT Enterprise that targets large organizations and offers an array of built-in cybersecurity controls. 

Image: OpenAI