‘The most stunning demo I’ve ever seen in my life’: ChatGPT impressed Bill Gates by acing AP Bio exam::ChatGPT’s AP Biology score demonstrates near human-level competency, according to tech billionaire Bill Gates.
Catch me surprised. A learning algorithm that may have tons of AP bio study material can ace the AP bio exam? Crazy…
Sure, it is obvious now, but less than a year ago barely anyone suspected that we had LLM more advanced than the Google assistant and Siri, let alone one capable of earning a medical degree.
Passing AP bio is a long way off from an MD, it’s just intro college bio. Still, the first part of the comment is fair, I think
oh I see, my bad.
LLM would tank in rounds.
This is the problem with standardized testing, it just doesn’t challenge our AI students and forces trainers to focus on the exam questions rather than an understanding of the material
Ok, the demo actually took place a year ago when GPT was smart. That makes more sense.
Gpt-4 helped me prepare my client for cross examination last week.
It predicted ten key questions he would likely be asked, and specified what he should say and what he mustn’t say, based on the way courts have decided cases historically in similar contractual disputes.
All of those questions came up. So yeah, GPT-4 is still smart.
You think all those questions came up because the other side is also using chatGPT?
I think it’s more likely the result of working through 200 prompts in five different chats using 5000-word templates and uploading hundreds of pages of pleadings.
It’s not like ChatGPT actually knows these things. It’s essentially a search engine. So if it aces an AP biology exam, it basically looked up the answers. It does a great job of making sense of data and putting it all together, but don’t be fooled into thinking that it knows everything about everything.
Sometimes that’s all that’s needed. Students, for example, often cram for an exam, regurgitate what they’ve memorized, and then promptly forget everything after the exam.
But what ChatGPT does is more like an open book test, rather than cramming before a test.
It’s great for doing tasks like these, yup.
