If you haven’t been following recent developments in AI, as reflected in programs like GPT-3, Dall-E and Stable Diffusion, you will be amazed at what they can do. And even if you have been following, you will be amazed at the newest product, which dropped yesterday. It’s a chatbot called ChatGPT by OpenAI, and you can go to its website and ask it sample questions. Yesterday I asked it a bunch of questions about topics that I write about. Here are some sample questions and answers, which are mindblowingly good. This AI can't quite do my job yet, but it's getting frighteningly close.
Each of the answers below were provided instantly. Yes, they are real. If I was giving them grades, most of them would get a solid B+. Here are some examples, organized by topic, starting easy and getting harder:
Biomechanics
B+. Points subtracted for being too wordy.
A!
An easier question with an apparent mistake about the function of the TFL:
Training
A!
B +. Greg Lehman might disagree, but that’s only because his chat algorithm will always disagree (even when asked about his own opinions.)
One for Adam Meakins:
Correct.
Diagnosis
Pain science
Some tough questions here:
Amazing.
Posture
Some “easy” questions followed by tough follow ups. Notice that the AI follows the thread of the conversation:
Hmm.
Conclusions
Some of the answers here are very good, some less good. Some just copy pasting conventional wisdom you find on WebMD, others seem to display actual reasoning. Some look like my own writing!
My impression is that this is amazing, frightening, and will cause some rapid changes in the world pretty soon.
Obviously, this raises a huge range of profound questions that I won't even be able to start addressing in this post. But we all better start asking good questions and trying to answer them soon.
What do you think? Let me know in the comments.
Picture of Hal9000 courtesy of Wikimedia Commons
Thanks for sharing this, Todd.
I've been reading about, and thinking a lot about, AI stewardship. There have been many calls to awareness by folks in the scientific community over the past decade plus who have expressed their concern (read: alarm) over the myriad of ways that this tech could impact our lives disastrously without ethical oversight. Can we entrust the determination of Truth, capital T, to those programming the AI? If the programmers are unethical, can the tech/algorithms be trusted? So many questions follow this line of inquiry that it makes the head spin.
That said, the examples you shared are compelling, a great example of what things could look like...
Thanks Todd! Really fascinating how its responses to the PP questions were highly nuanced and line up nicely with current discussions on the topic, yet the questions on posture were slightly less sophisticated. Getting more advanced, but not yet to the level of the Android known as Greg Lehman.