OpenAI Unveils GPT-4 With Higher Test Scores Despite Ongoing Flaws
OpenAI launches GPT-4, a multimodal AI upgrade that scores in the top percentiles on major standardized exams but still struggles with fact-inventing hallucinations.
OpenAI releases GPT-4, a highly anticipated upgrade to the technology behind ChatGPT, marking the latest move in a fierce race among tech giants to dominate artificial intelligence. This new multimodal model accepts both image and text inputs to generate text, offering a significant step forward from the previous text-only capabilities.
The updated model shows impressive performance improvements, scoring in the 88th percentile or above on the LSAT, SAT Math, and SAT Reading and Writing exams. Unlike GPT-3.5, which landed in the bottom 10% on a simulated bar exam, GPT-4 achieves a score in the top 10% of test takers and handles much more nuanced instructions.
Despite these advancements, OpenAI warns that GPT-4 still suffers from key limitations, most notably its tendency to hallucinate or invent facts. CEO Sam Altman cautions that the technology remains flawed and limited, urging users to take great care when relying on the system for complex tasks.