GPT-4 passed the Uniform Bar Exam (multiple-choice and written components), scoring in the 90th percentile relative to human bar-takers. This represents a sharp advance from GPT-3.5, which scored in the bottom 10th percentile on the same exam. The source indicates this is the first LLM capable of powering professional-grade legal products, though human attorney oversight remains necessary. The development signals that domain-specific legal tasks now fall within LLM capability rather than outside it.
258 other tracked developments touching the same claims.