OpenAI’s recent update to ChatGPT introduces impressive new image analysis capabilities. In this article, we’ll guide you on utilizing these features effectively while keeping certain considerations in mind.
OpenAI has recently rolled out a substantial upgrade to ChatGPT (available on both Apple and Android devices), introducing two noteworthy features: AI voice options for more immersive interaction with the chatbot and image analysis capabilities, which bear resemblance to the existing features found in Google’s Bard chatbot.
After extensively testing ChatGPT’s capabilities, I can’t help but be both impressed and somewhat apprehensive about its potential. While the web browsing beta introduced through ChatGPT Plus was impressive, it also raised concerns about its impact, especially for online content creators. The introduction of the new image feature for OpenAI’s subscribers has left me with similarly mixed feelings.
While I haven’t had the opportunity to explore the new audio capabilities yet (other colleagues have), I did get a chance to test the upcoming image features. Here’s a guide on how to utilize ChatGPT’s image search and some tips to get you started.
How to Access ChatGPT’s Image Features
The update is expected to be rolled out before the end of the year, but the exact release date for the image and voice features remains uncertain. As with most OpenAI updates, paying subscribers will have the first access.
There are three ways to upload photos in the ChatGPT mobile app:
Camera Capture: You can select the camera option to the left of the message bar and take a fresh photo with your smartphone. Before uploading the image, you can use your finger to draw a circle around the subject you’d like the chatbot to focus on.
Device Photos: You can also choose photos from your device to upload and select files saved on your phone.
Desktop Upload: Desktop browser users of ChatGPT can upload saved photos directly from their computers. Although there’s currently no option to upload videos to the chatbot, you can submit multiple images in a single prompt.
Advice for Experimenting With the New AI Tools
While computer vision technology has been available to the public before, the combination of accessibility and a powerful chatbot suggests that something unique and potentially transformative might be happening here. However, a few precautions are in order:
- Never upload personal or sensitive photos to ChatGPT as you experiment with the image feature.
- To limit how long OpenAI retains your data and AI interactions for training purposes, go to Settings, then Data Controls, and disable Chat History & Training. This will result in your information being deleted after one month. Remember to do this individually for every browser you use to access ChatGPT, both on PC and mobile.
For the best results with ChatGPT’s image analysis, consider the following:
- Upload clear and well-lit images for better accuracy. While ChatGPT may occasionally make errors, it can accurately identify objects such as an orchid plant, international coins, a charging cable, and even a Steve Irwin Funko Pop in my apartment.
- Exercise caution in accepting its answers without verification. ChatGPT mistakenly identified my daily multivitamin as an erectile dysfunction pill.
ChatGPT does have its limitations. When presented with a random mural photo, it couldn’t identify the artist or location. However, it had no trouble recognizing various San Francisco landmarks, such as Dolores Park and the Salesforce Tower. While it might still seem a bit gimmicky, ChatGPT’s image analysis feature can be enjoyable for anyone exploring a new city, country, or neighborhood.
OpenAI has implemented a significant safeguard regarding this new feature: the chatbot’s ability to answer questions that identify humans is limited. The chatbot prioritizes user privacy and safety, restricting responses related to real individuals based on images, even if they are famous. While ChatGPT may not entirely refuse to answer questions related to explicit content, it shows hesitation in providing specific descriptions of adult performers beyond explaining their tattoos.
It’s worth noting that during one conversation, I had a chatbot version with the image feature seemingly circumventing OpenAI’s safeguards. Initially refusing to identify a meme of Bill Hader, the chatbot incorrectly guessed that an image of Brendan Fraser in “George of the Jungle” was Brian Krause from “Charmed.” Upon further inquiry, it eventually corrected itself.