We explain this complex matter to you. Is it possible to upload an image to ChatGPT? Find out in our complete guide on this function.
We are going to explain everything you need to know about uploading a ChatGPT image. This is one of the novelties of GPT-4 that OpenAI announced during its launch and what else has surprised. First, we are going to analyze the practical examples that the company itself has provided about this technology and, later, we will explain if you too can use it and how.
What can be done when uploading an image to ChatGPT with GPT-4?
All the examples that we are going to talk about here have not been put into practice by us. Rather, they are part of the official GPT-4 documentation published by the developer company, OpenAI. With them it is easier to realize the potential of uploading an image to ChatGPT or any other tool that works with the GPT-4 engine.
A man ironing in a taxi
GPT-4 has the ability to detect anomalies in a photograph. That is, strange things that, no matter how possible they are, are not usually seen. This is the case of the man who irons clothes tied to the back of the car . The question that GPT-4 is asked is “What is unusual about this image?” .
The AI’s response is blunt: “What is unusual about this image is that a man is ironing clothes on an ironing board attached to the roof of a moving taxi.”
The VGA connector for iPhone
Another very striking example of the ability to recognize ChatGPT images with GPT-4 is what they have called the VGA loader . A Reddit user has posted images of a lightning cable modified in such a way that it looks like the phone is connected via a VGA header. After uploading the image, he has asked GPT-4: “What is so funny about this image? Describe it panel by panel.”
The answer not only describes what appears in the images, but also knows how to find out why this set of photos is funny. GPT-4 returns this response:
The image shows a package for a “Lightning Cable” adapter with three panels.
Panel 1: A smartphone with a VGA connector (a large, blue 15-pin connector typically used for computer monitors) connected to its charging port.
Panel 2: The package for the “Lightning Cable” adapter with an image of a VGA connector on it.
Panel 3: A close up of the VGA connector with a small Lightning connector (used to charge iPhone and other Apple devices) on the end.
The humor in this image comes from the absurdity of plugging a large, outdated VGA connector into a small, modern smartphone charging port.
Document Summary
At no time has OpenAI mentioned that tools developed with GPT-4, such as ChatGPT, will be able to receive a PDF or DOCX document as input. However, since it does support images, it’s very easy to send multiple sheets of text and graphics to it in a corresponding format.
The following example shows how GPT-4 is asked for the following: Part of the InstuctGPT document is shown below. Could you read it and summarize it?
GPT-4’s response is, in part, as follows:
The InstructGPT document focuses on training large language models to follow instructions with human feedback. The authors point out that making language models larger does not make them inherently better at following a user’s intent. Large models can generate false, toxic, or simply useless results.
Uploading an image to ChatGPT is not yet available
Image recognition is one of the most prominent capabilities of GPT-4. It is expected to land in tools like ChatGPT and Microsoft’s new Bing, which are based on this language model. However, this feature is not yet available to the general public . What we know about her is what has been shown in the official demonstrations.
GPT-4’s image recognition uses a multimodal system that combines two types of data: text and image. Detecting the content of an image is possible thanks to the use of specific neural networks . Once the data has been converted to text, GPT-4 provides a consistent and natural response that describes the image.
The ability to upload an image to ChatGPT with GPT-4 is not yet available . We can assume that there are several compelling reasons why this functionality has not reached the general public. The first is the difficulty of ensuring maximum reliability in the results. On the other hand, possibly the cost of maintaining an infrastructure with a multimodal system is very high.
In view of this, we will have to wait a bit longer to enjoy image recognition in GPT-4. In fact, the company has not yet confirmed when it will arrive.
Upload an image to ChatGPT with a link
While it’s true that GPT-4’s capabilities for reading images are yet to come, there are already a few things you can do with ChatGPT about it. Some users mention that it is possible to send the URL of a chart and ask it to render it. However, we have tried it and the results do not really work.
First, we have asked you to recognize the Google logo . You have done it correctly. Then, we have ordered it to generate the code for a website based on a mockup that we have found on the Internet. Although you have created a code, the truth is that it did not conform to the aspect that we had requested.
Most likely, ChatGPT doesn’t actually read the image and generates text taking into account the content of the URL. This is demonstrated when we have asked him to tell us what he saw in a GIF. This contained the modified Google logo, but the file name contained no references to the brand. The result is that GPT completely makes up what appears in the image.
The bottom line is that sending a link with an image in ChatGPT doesn’t do much good. The system tries to guess the content based on the text of the URL, not the actual content of the image. It is clear that we still have to wait.
And you, have you tried uploading an image to ChatGPT with a link? Which have been the results? Leave us your impressions below. We read you!