OpenAI Unveils GPT-4; Outperforms 90% of Humans on Toughest Exams

OpenAI has been revolutionizing the world of artificial intelligence since its inception. In 2022, it unveiled the GPT-3 model, quickly recognized as a breakthrough in the field.

However, OpenAI continued. The company has continued to push the boundaries of AI and is now rolling out its latest model, GPT-4. Experts studies show this new model outperformed 90% of humans in the world’s toughest exams.

The CEO of OpenAI, Sam Altman, claims that GPT-4 is a crown of innovation equipped with more complex directives than its predecessor GPT-3. Altman also emphasizes that GPT-4 is more dependable and versatile than the previous version, which was already a breakthrough in the field.

The unveiling of GPT-4 was geared towards displaying multiple capabilities and superiority over its predecessors. The model’s improved performance and new features are expected to significantly impact the field of artificial intelligence and natural language processing.

Outperformed America’s Toughest Tests

At the demonstration, the engineers were feeling confident, and rightfully so. GPT-4 exceeded expectations by outperforming 90% of humans in the world’s toughest exams. According to multiple sources, the model achieved an impressive 93% on the SAT exams, including over 85% on the math tests.

It also performed exceptionally well on other exams, achieving over 88% on the LSAT and 80% on the quantitative GRE. The model’s performance on the GRE verbal was even more impressive, hitting a staggering 99%.

Can Handle Image Input

In addition to its impressive exam performance, there is even more, to be excited about with GPT-4. Developers have enhanced the model to handle images, increasing its interaction capabilities. This significant development sets GPT-4 apart from other users in the search space.

The ability to process images has yet to be added to the market, and it can save a lot of time for users who often have to research text and images separately. OpenAI has provided a great example of this new capability on its website.

They showcase an illustration of a user giving the chatbot an image of baking ingredients and asking what they can be used to make. While the video aspect of the bot could be clearer, this development shows the potential for a more interactive and efficient experience in the future.

GPT 4 Advancements

According to OpenAI, GPT-4 has been trained with a new approach that includes human feedback, allowing continuous improvements. In addition, the company has collaborated with over 50 experts, including those in AI safety and security, to get feedback and ensure responsible development.

The media has reported that OpenAI is partnering with many companies, such as Duolingo and Khan Academy, to integrate the new model into their apps and products. Khan Academy, in particular, plans to use the technology to create automated tutors, with GPT-4 expected to pave the way for automation in this field.

The potential profitability for these apps is significant, given that the chat GPT-4 training data reportedly includes over 174 billion parameters, compared to GPT-3 or ChatGPT’s 175 million. This difference is a massive leap forward in capabilities, and the possibilities could be beyond what we can imagine now. While the release date is uncertain, many sources predict it will be later in 2023, with tech enthusiasts eagerly awaiting this tech marvel.