Can OpenAI’s latest language model outsmart its predecessor?


GPT-4, the latest AI language model from OpenAI is now available for ChatGPT Plus subscribers and can be accessed via the new Bing.

OpenAI has unveiled its latest language model, GPT-4, which powers Microsoft’s new Bing and several third-party apps.

With the new version, OpenAI claims that GPT-4 is smarter and more sensible than GPT-3.5 and the previous iterations and is trained on Microsoft Azure AI supercomputers with NVIDIA H100 GPUs.

How different is GPT-4 from GPT-3.5, which we have been using for the last few months? Here is a comprehensive comparison between the two.

ChatGPT-4 can accept an image as an input

One of the significant differences is that GPT-4 can accept images as input, allowing users to make queries with advanced reasoning capabilities.

GPT-4 can recognize and analyze the image and build responses around it, while GPT-3.5 can only accept textual input.

GPT-4 can generate longer responses

Another improvement is that GPT-4 can generate longer responses, with up to 25,000 words, while GPT-3.5 was limited to 8,000 words. This allows for more detailed copies and responses on GPT-4-based apps or services.

GPT-4 can clear competitive exams

GPT-4 is also smarter than its predecessor, scoring in the 90th percentile on a uniform bar exam compared to GPT-3.5’s 10th percentile.

Additionally, GPT-4 produces 40% more factually correct responses and is 82% less likely to produce disallowed content on request.

GTP-4 is not free and has limitations

However, GPT-4 is not free, and currently, only ChatGPT Plus subscribers can access it. Some apps like Stripe and Duolingo are already using the latest language model.

It is essential to note that GPT-4 still has limitations, including social biases, hallucinations, and adversarial prompts. Therefore, it should be used with caution.