OpenAI'S GPT-4 Finally Gets IMAGES (Now RELEASED!)

GPT-4 with Images: Overview

This transcript discusses the recent addition of images to GPT-4, a language model developed by OpenAI. The speaker explains that this feature was announced earlier in the year and is slowly being rolled out to some users. The transcript provides examples of how GPT-4 with images works and its potential applications.

Introduction

• Yesterday's video discussed the addition of images to GPT-4.

• This feature was announced earlier in the year during a developer live stream.

• GPT-4 with images is slowly being rolled out to some users.

Visual Inputs with Bing

• Microsoft recently rolled out visual inputs with Bing to around 5% to 2% of users.

• Users can access this feature by clicking on a special icon in the Bing chat box.

• This feature allows users to add an image, upload from a device, or take a photo.

User Examples

Understanding Images

• A Twitter user used an image from Reddit to test GPT-4's understanding capabilities.

• The AI was able to recognize what the chord in the image was and even identified a Dragon Ball Z sticker on it.

Captcha Test

• Another example involved solving a captcha test using GPT-4.

• The AI recognized that it was a captcha and quickly solved it.

Applications of GPT-4 with Images

• There are many potential applications for GPT-4 with images, including helping people solve computer issues, recognizing objects in photos, and solving captcha tests.

Bing's Visual Input Capabilities

This section discusses the visual input capabilities of Bing's AI system, GPT 4. It showcases how GPT 4 can recognize and describe images accurately.

Recognition of Nephron Image

• Bing recognizes an image of a nephron accurately, describing it as a basic unit of the kidney that filters blood and produces urine.

Identification of Cross-Section Tissue Image

• Bing identifies a cross-section tissue image correctly, stating that it is most likely a muscle tissue.

• When prompted further, Bing states possible signs of disease based on the image alone.

Change in Response to VGA Connector Image

• The transcript highlights a change in GPT 4's response to an image of a VGA connector plugged into a phone.

• The original response was more accurate than the current one, which only describes what it sees without understanding the humor behind it.

Conclusion

This section concludes the discussion on Bing's AI system, GPT 4. It emphasizes its potential applications and capabilities in recognizing and describing images accurately.

Potential Applications for Images Recognition

• The transcript highlights that the application for recognizing images is going to be unbelievable.

• The conclusion emphasizes GPT 4's potential capabilities in recognizing and describing images accurately.

GPT-4's Potential Impact on the Medical Field

The speaker discusses how GPT-4's capabilities in identifying medical conditions through visual inputs can potentially impact the medical field.

GPT-4 vs. Doctors

• GPT-4 has the potential to outperform doctors in identifying medical conditions due to its ability to be trained on millions of images of a specific type of condition.

• The speaker believes that if GPT-4 is fine-tuned for certain medical capabilities, it could have a significant impact on the medical field.

Visual Input Examples

• The speaker provides examples of GPT-4's visual input capabilities, such as identifying an image of a man ironing clothes on an ironing board attached to a roof of a moving taxi and explaining a meme about different countries represented by chicken nuggets.

• The speaker notes that GPT-4's visual input goes beyond simple image identification and includes context recognition, which is different from early models.

Rollout Plans for GPT-4

The speaker discusses rollout plans for GPT-4 and speculates about how it will be implemented.

Chat vs. Bing

• Currently, there are 100 million users every day using chat GPT, but it is unclear whether or not images will be rolled out in Bing before being added to chat.

Implementation Details

• It is unknown how exactly images will be implemented into chat GPT or if Discord will continue to be used for AI applications.

• The speaker speculates that GPT-4 is being tested slowly to avoid issues on a full-scale platform.

GPT-4's Potential in Website Development

The speaker discusses how GPT-4 can be used in website development.

Hand-drawn Mock-up

• The speaker demonstrates how he was able to take a hand-drawn mock-up of a joke website, take a photo of it, and give it to GPT-4, which was able to code it within seconds.

• The speaker notes that there are many potential applications for GPT-4 and that they are still figuring out new ways to use it.