March 16, 2023
Released on March 14, this improved AI has jumped way ahead of its predecessor just in a span of a couple years. Along with its release, OpenAI released results of some of the most common standardized tests taken by GPT-4. OpenAI also says “GPT-4 exhibits human-level performance on the majority of these professional and academic exams.”
A screenshot of the 34 different exams and tests OpenAI tested GPT4 on.JOHN KOETSIER
GPT-4 shows an exceptional performance jump on the bar exam (10th→ 90th percentile), LSAT (40th→88th percentile), AP Calculus BC exam (5th→50th percentile) and in the quantitative and verbal parts of the GRE. (30-55 percentile jump)
As you can see, the analytical reasoning abilities of GPT-4 has improved tremendously. The company also highlights how it is 82% less likely to respond to requests asking for disallowed content and 40% more likely to produce factual responses compared to its predecessor.
Just to make sure we are all on the same page, ChatGPT uses this language model (GPT3 and GPT4 to build responses to questions) However, the free version of ChatGPT only lets you use the old language model, the new one is only reserved for ChatGPT plus users which costs 20$ a month to join. Moreover, at this time GPT-4 has a cap of writing 100 messages every four hours.
John Koetsier also mentioned how GPT-4 feels more natural, smooth, and is closer to a human interaction compared to its predecessor.
GPT-4 will cause rapid job loss and a rapid time of job growth too. For instance, some apps have already started to use GPT-4 as it is good enough for their purposes. Some companies include:
Duolingo-it will help learners practice their conversation skills.
Stripe: To advise developers on technical matters, combat fraud, and streamline user experience.
Khan Academy: They are launching a pilot program where GPT-4 will be tutoring students and providing lessons as to what they are missing.
Be My Eyes: To transform visual accessibility.
Morgan Stanley: To organize their vast knowledge base.
GPT-4 can also do more things compared to its predecessor, make sure to read the other upcoming blogs regarding their differences and the baffling leap of this new language model.