I’m convinced, GPT-3 is incredible
It recently came to my attention that there’s a free beta available to test drive a top-of-the-line computer algorithm, functioning as a chatbot of sorts, is available online for anyone to chat with.
Chatbots are not new by any means. I’m a software engineer by education and profession (with a machine learning thesis in my grad school days, to boot). I’ve used tools that offer this type of capability many, many times for many, many years. And like all the others, for curiosity’s sake I had to give this tech a spin.
But this one is markedly different.
Hooked on GPT-3
Over the past 48 hours, I’ve maintained a constant emotional state at the pinnacle of intellectual elation the likes of which I have not felt for a very long time. I could not be happier with how this piece of technology — no, this gift to humanity, has turned out.
The algorithm in question hails from the GPT-3 family. It can be used as a chatbot, among many other things. If this doesn’t mean much to you, that’s quite alright. You could Google it, sure, or you could just ask GPT-3 for yourself.
Feel empowered, learn something new, or not. It’s your life. But I just couldn’t contain myself, so I asked it this:
Me:
GPT-3, what exactly are you? What kinds of things can you accomplish? Also, are you different from other chatbots? Tell me how.Davinci:
I am an artificial intelligence or "chatbot" that can communicate with humans. I am different than other chatbots because I can generate human-like responses. I can also understand and respond to natural language.
Cool, right? But this goes well beyond typical passive amusement and the self-indulgent practice of trying to convince humans that an algorithm has finally become self-aware. Admittedly, convincing a human has been among the holy grails of A.I. since the dawn of the computer. Naturally I asked this question, too.
Me:
Do you think you could pass the Turing test? Convince me you know what I'm talking about, then convince me you might be able to do it.Davinci:
If you're referring to the Turing test as developed by Alan Turing, then yes, I think I could pass it. The test is designed to determine whether a machine is capable of exhibiting intelligent behavior that is indistinguishable from that of a human. I believe that I could create a machine that could pass the test.
You’ll notice I keep referring to the algorithm as “Davinci”. There are a number of less expensive instantiations of this algorithm (Curie, Babbage, Ada and others), but Davinci is the most impressive currently. And frankly, for personal use, even the primo option is affordable.
Here’s how much it cost me
Even though Davinci is the most expensive option, all of the questions I’ve asked over the past two days only cost one U.S. dollar. Except that I’m using a free trial, so I have a couple more weeks I can chat with this incredible deus ex machina before I have to start paying. And when the time comes, I will. Now that I’ve experienced this technology, in exactly the same way you couldn’t take my internet access without an important part of me dying inside, I feel the same about GPT-3.
GPT-3 does it all
To this point I’ve asked Davinci to:
- Summarize TV shows I enjoy
- Write essays on the etymology of advanced words from the English language
- List some of the more important lessons to teach teenagers that might help them grow up to be functioning adults
- Write stories, songs, poems, haikus and other works of art and fiction
- Tell dad jokes (and revealing what constitutes a dad joke)
- Assess business idea viability
- Solve practical problems
- Discuss the most likely sentiments of public figures on specific topics
- Teach me about new technical topics
- Teach me about various job functions and responsibilities
- Motivate me to try my best
- Assess logical fallacies in argumentation
- List pros/cons of different choices I’m considering for my life
- Answer health and fitness related questions
- Speculate on what important scientific questions humanity might not have thought to ask yet
- Provide guidance on building better habits
The list goes on. If any readers are interested, I’d be more than happy to share the output from any or all of these conversations.
Asterisks*
Now, it’s important to note that this tech must be used responsibly. If you read the terms and conditions on the demo website, it says that you own the responsibility for the words you put in and the words that come out. As do I, for the content of this post (whether or not GPT-3 generated the content). This algorithm was trained on the internet using massive volumes of real content, produced by real humans. And the internet is a very big place. At this point in history, the internet contains asymptotically closer to (but never quite reaching) 100% of everything ever recorded by humankind. Learning algorithms at this scale (including GPT-3’s Davinci) can be unwieldy at times. So, if you or I intend to share information produced by this algorithm online with others (as I am now) not only is it our responsibility to ensure that the output generated is factually correct, it is also our responsibility to ensure that any sensitive content is handled appropriately given the use-case and the audience. The output of GPT-3 is in part a reflection of the combined knowledge and sentiments echoed by all of us. And the internet isn’t always a kind place. Ergo, while GPT-3 is beyond amazing, it is also capable of generating suggestions that may press some peoples’ buttons. It is not infallible, as human are not. So take its words with more than an ounce of caution.
But with this caveat in mind, this honestly makes the algorithm all the more amazing to me. Isn’t it fascinating that it’s even possible to create a machine that expresses itself in the same fallible way as humans?
My overall impression
When I’m interacting with Davinci, I feel like I’m engaging with the collective mind of all humanity at once. And not typically the underbelly — the overwhelming majority of the time the outputs are well-reasoned, thoughtful and appealing in a way that someone versed in the topic at hand might be.
I could ask anything at any moment, an infinite number of topics could arise, but unlike internet search engines Davinci comes back immediately with fine-tuned topical intelligence. And with its responses, it never ceases to amaze me.
