As the world’s largest artificial intelligence conference, NIPS (Neural Information Processing Systems), kicks off today in Long Beach, California, tech bloggers and insiders are anxiously awaiting the latest announcements from OpenAI, the research lab founded by Tesla CEO Elon Musk and several others.
And they did not disappointed.
OpenAI announced the general release of their new GPT-4 model, which they say exhibits “human-level performance” on a number of professional benchmarks, including the natural language processing task of question answering.
This is a stunning development and cements OpenAI’s position as the leading AI research lab in the world.
But what does it really mean?
First, some context. OpenAI first announced their GPT model, which stands for “generative pre-trained transformer,” last year at NIPS. At the time, they showed that their model was able to generate remarkably human-like text.
While the original GPT model was impressive, it was not perfect. For example, it often generated text that was grammatically incorrect or made little sense.
But with the release of GPT-4, OpenAI says they have largely solved these problems.
In addition, GPT-4 is much larger than the original GPT model. It is trained on a whopping 10 billion parameters, compared to just 1.5 billion for the original GPT.
This increased size has resulted in a model that is not only more accurate, but also faster and easier to train.
OpenAI says that their new model can be trained on a standard NVIDIA GPU in just a few hours, compared to days or weeks for the original GPT.
But the real test of any artificial intelligence model is how it performs on real-world tasks.
And by this metric, GPT-4 is truly impressive.
OpenAI says that their new model can beat 90% of humans on the SAT, a widely used college admissions test.
Furthermore, they say that GPT-4 is also effective at tasks such as translations, summarization, and even creating code.
In particular, OpenAI says that their new model can generate code that is syntactically correct and runs without errors more than 95% of the time.
This is an incredible achievement and shows just how far artificial intelligence has come in recent years.
Of course, GPT-4 is not perfect. For example, it still makes occasional mistakes in its generated text.
But overall, it is a truly impressive achievement and a major step forward for artificial intelligence.