MiniGPT-4 arrives totally free: Do not pay to use Chat GPT-4

Is Chat GPT-4’s reign coming to an end? A new open source project named MiniGPT-4 is gaining attention as it offers comparable performance to Chat GPT-4. Developed by a group of students, this AI technology is now available for free demo access. MiniGPT-4’s efficiency in complex image recognition processes has caused a sensation among AI enthusiasts.

Chat GPT-4 is still the most widely used AI technology globally and has gained immense popularity. Its applications range from creating web applications to resolving various doubts or even in video game programming. However, MiniGPT-4 aims to provide a more efficient option on a smaller scale, with the added advantage of being free and not limited to simple processes like generating fake faces with AI.

AI Robot

Created by a group of students

The MiniGPT-4 project was initiated by a team of students from King Abdullah University of Science and Technology, Saudi Arabia, with the aim of providing a technology that is similar to Chat GPT-4, but with a unique approach and system. Moreover, the team wanted the project to be an open-source initiative, freely accessible to everyone.

The MiniGPT-4 project relies on advanced technology to power its functionality, including the Vicuna language decoder based on LlaMA and the powerful BLIP-2 component, which enables sophisticated image recognition capabilities that can be leveraged to answer questions based on visual data. Despite being dubbed as a “mini” version of the GPT-4, MiniGPT-4 boasts an impressive 90% efficiency level, according to GPT-4’s analysis. While it may not quite match the capabilities of OpenAI’s AI, the performance of MiniGPT-4 is certainly noteworthy and has garnered significant interest from the tech community.

What is he capable of doing?

To put it simply, MiniGPT-4 has a similar capability to Chat-GP4 in image recognition, but with a 10% difference in its success rate. The project managers have provided clear examples to showcase the AI’s visual analysis capacity. One example involved analyzing a photo of a sick plant to test if the AI could accurately detect what was happening to it.

According to reports, MiniGPT-4 provided precise results with just one question and the image of the plant in question. Another example involved uploading the Vicuna logo to ask MiniGPT-4 for its opinion, and the AI provided a complete description and commentary on each element, concluding with a final assessment.

Despite its impressive capabilities, MiniGPT-4 still has room for improvement. The AI’s responses can be repetitive and rely too heavily on certain phrases and expressions, especially when analyzing images from public sources. However, the developers acknowledge that this issue can be solved with additional training. Currently, MiniGPT-4 can analyze photos and generate marketing descriptions, songs, and food-related comments.

The current version of MiniGPT-4 has undergone a 10-hour learning process with the help of four A100 GPUs and 5 million text and image pairs. With further improvements, the AI has the potential to exceed its current performance. Interested users can test MiniGPT-4 on its Github website.

To use MiniGPT-4, simply visit its website and upload the image you want to use for the conversation. However, keep in mind that if the photo is very large, it may take some time to upload. After that, you can type your comment or question and wait for the response. It is worth noting that the response speed may vary and sometimes require up to 50 seconds to generate a reply, as we experienced during our test. Nonetheless, the waiting time is displayed on the website to manage expectations.