OpenAI, a company owned by Microsoft, unveiled GPT-4 yesterday, March 14. GPT-4 is it's latest multimodal model that can process both image and text inputs. "We've created GPT-4, the latest milestone in OpenAI's effort in scaling up deep learning," the company said in a blog post on Tuesday, March 14.
In addition, OpenAI has published a roster of assessments that GPT-4 has completed, along with the corresponding scores, which are impressive.
Notably, the language model achieved an 88% on the LSAT (Law School Admission Test) and 89th percentile on the SAT (Scholastic Assessment Test) Maths exam. Furthermore, it received an 80th percentile on the GRE Quantitative exam and scored the 99% and 54% on the GRE verbal and writing exams, respectively.
According to OpenAI GPT-4 has shown superior performance to ChatGPT, achieving higher approximate percentile scores across various assessments. For instance, on the Uniform Bar Exam, ChatGPT obtained a percentile score of 10, while GPT-4 achieved a percentile score of 90. Additionally, in the Biology Olympiad, ChatGPT scored in the 31st percentile, whereas GPT-4 (with vision) scored in the 99th percentile, further highlighting its capabilities.
"We've spent 6 months iteratively aligning GPT-4 using lessons from our adversarial testing program as well as ChatGPT, resulting in our best-ever results on factuality, steerability, and refusing to go outside of guardrails." GPT-4, the newly developed AI model, is deemed to be more dependable, innovative, and proficient in processing intricate instructions than its predecessor, GPT-3.5. It surpasses other existing large language models (LLMs), including the most advanced state-of-the-art (SOTA) models that may have been constructed based on specific benchmarks or additional training methods as reported by IANS.
