All eyes on GPT-4: Bigger. Better. Brainier.

In the AI arms race, OpenAI has emerged with the big guns; GPT-4 is here and it might just have the potential to turn ChatGPT into a 'predecessor'.

With ChatGPT, we saw the inception of GPTZero. Little do we know what is yet to come!

What has been introduced?

This is the latest advancement by OpenAI in its efforts towards elevating artificial intelligence and deep learning. Their new GPT-4 is a large multimodal model that can now also accept image inputs along with text, to give text outputs.

The new AI model is available starting 14th March 2023 in OpenAI's API (with a waitlist) and for ChatGPT+ users (of course, with a usage cap).

How is this different from ChatGPT?

This platform is aiming to offer an image input capability where text outputs will be generated for provided photos. The AI-language model has been upgraded to exhibit human-level performance on various professional and academic benchmarks.

GPT-4 can solve difficult problems with greater accuracy, thanks to its broader general knowledge and problem solving abilities.

GPT-4 can generate up to 25,000 words which are approximately 8 times more text than ChatGPT. The model is trained till August 2022 and keeps a check for adversarial usage, unwanted content, and privacy concerns.

A live stream hosted by  Join Greg Brockman, President and Co-Founder of OpenAI showed how GPT-3.5 was not able to process complex commands very efficiently. The same prompts when used on GPT-4 provided mind-blowing outputs!

Head over to the video below for a developer demo showcasing GPT-4 and some of its capabilities/limitations.

Find out more about what has improved in GPT-4 on OpenAI's website.

Why is this better?

In an effort to understand the difference between the two models, the team at OpenAI tested on a variety of benchmarks, including simulating exams that were originally designed for humans.

The team experimented with some common exams like the Olympiads and AP free response questions or by purchasing 2022–2023 editions of practice exams and there was no specific training for these exams.

We also evaluated GPT-4 on traditional benchmarks designed for machine learning models. GPT-4 considerably outperforms existing large language models, alongside most state-of-the-art (SOTA) models which may include benchmark-specific crafting or additional training protocols.
Source: OpenAI

The results were significant, for example, it passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT-3.5’s score was around the bottom 10%. The distinction between GPT-3.5 and GPT-4 can be subtle; when the complexity of the task reaches a certain threshold - GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than its previous versions.

For more statistics on the performance of GPT-4, head over to this report.

Where have you used this before?

Microsoft's Bing bot had been running the GPT-4 language model all this while and we were all unaware.

“We are happy to confirm that the new Bing is running on GPT-4, customized for search,” according to a blog post from Yusuf Mehdi, Microsoft’s head of consumer marketing. “If you’ve used the new Bing in preview at any time in the last six weeks, you’ve already had an early look at the power of OpenAI’s latest model.”

To sum up, Microsoft with OpenAI is preparing to open fire and only time will tell how the rivals fare in this fierce battle of artificial intelligence!

Stay up to speed with the latest in AI! Check out more on CohesiveAI's blog.

Sources: The Time of India, Eluna.ai, OpenAI