What Makes OpenAI’s New GPT‑4 a Game‑Changer for Multimodal AI?

OpenAI announced the launch of GPT‑4, the latest model in its AI language series, capable of processing both text and image inputs and powering applications such as ChatGPT and the new Bing.

The company claims the model is more creative, collaborative, and better at solving problems than previous versions, though it still suffers from hallucinations and can generate harmful content.

GPT‑4 is already integrated with platforms like Duolingo, Stripe, and Khan Academy, and is available to the public via ChatGPT Plus ($20/month) and an API for developers.

Compared with GPT‑3.5, GPT‑4 shows subtle improvements in everyday conversation, and its performance on benchmarks such as the Uniform Bar Exam, LSAT, SAT Math, and SAT Evidence‑Based Reading & Writing is notably strong.

OpenAI notes that the system underwent six months of safety training, reducing disallowed‑content responses by 82% and increasing factual accuracy by 40% relative to GPT‑3.5.

Despite the advances, OpenAI’s CEO Sam Altman warned that expectations are high and the model still has limitations.

编辑：高朋

Related reading:

GPT‑4, arriving next week!

OpenAI CTO Mira Murati: ChatGPT technology should be human‑centric

The world‑changing impact of ChatGPT and GPT‑4