OpenAI Launches “Human-Level Performance” GPT-4 Language Model

Larger Model with More Data and Weights

On Tuesday, OpenAI unveiled its latest version of the GPT-4 language model, which has “human-level performance” on various professional tests, according to the AI firm. The new ChatGPT-4 is “larger” than earlier versions, containing more weights in its model file, which makes it even more expensive to operate.

“Scaling-Up” In AI Advancements

To achieve better AI results, many researchers believe in “scaling up” through running increasingly larger models on thousands of supercomputers in training processes that can cost tens of millions of dollars. GPT-4 is an example of that approach.

Microsoft Azure Trains the Model

The new model was trained using Microsoft Azure, which has invested billions in OpenAI to date. The company didn’t specify the particular model’s size, citing competition details, nor the hardware utilized to train it.

Preview of Advancements in Consumer Products

GPT-4 powers many of the artificial intelligence demos showcased in the tech sector for six months, such as Bing’s AI chat and ChatGPT, of which the latest version offers new advancements to consumer products such as chatbots. Microsoft confirms Bing’s AI chatbot uses GPT-4.

Better Results and Standardized Tests Performance

The new model reportedly produces fewer factually incorrect responses, chats less about forbidden topics, and gains a better result than humans on many standardized tests. OpenAI claims that the new software scored at the 90th percentile in a simulated bar exam, the 93rd percentile on an SAT reading exam, and the 89th percentile on the SAT Math exam.

Known Limitations in New Software

However, OpenAI acknowledges that the new software is not yet perfect and emphasizes that it is less capable than humans in some scenarios. The GPT-4 still experiences “hallucination,” meaning it tends to make stuff up and is not always reliable. The software is still prone to insisting it is correct, even when it is not.

Available to Paid Subscribers

The new model will be available to paid ChatGPT subscribers and is also available as part of OpenAI’s API, allowing programmers to integrate the AI into their applications.

Credit: https://www.cnbc.com/2023/03/14/openai-announces-gpt-4-says-beats-90percent-of-humans-on-sat.html

ENND