Top 10 Experiments with GPT-3 – The Ponicode Selection
Open AI blesses us with a new GPT release and we have not been disappointed. After years of development, GPT-3 or generative pre-trained transformer 3 is now available to a selected audience. A few lucky developers are currently experimenting with the endless range of possibilities that this tool seems to offer.
Open AI was funded by Elon Musk and it aims to stir up the AI industry. If it’s not already doing that, at least it stirs up the hype around AI: GPT-3 has been getting a lot of attention for the seemingly unlimited range of possibilities it offers. GPT-3 extended its capacity to answer questions and solve problems. It understands language tones, non direct messages and other subtleties of the human speech. It still tends to make basic mistakes along the different experiments it is challenged with. Developers seem confident about the margin for improvement though. Given that GPT-3 is reaching this level now it means a very promising future once some fine tuning and improvements are executed.
It is way ahead of anything done before. Ten times more parameters than any similar programs were fed to it. All English Wikipedia pages were fed to it and they only represent 0.6% of its training data.
We were non stop talking about it here at Ponicode and we wanted to share the use cases that we were the most excited about in a mandatory Top 10.
We want to get started by a of couple experiments that show how A.I. will be able to assist professionals in their work and reduce the possibility of human error. In this experiment, the A.I. has been fed with patient files, describing their profile and symptoms in a few lines. The A.I. spontaneously makes suggestions of what the disease could be. GPT-3 got away with an impressive 8 out of 10 correct guesses. This could become an amazing support to doctors, a great tool to investigate and prompt follow up questions.
Wow effect: 10/10 When it comes to health and adding an extra brain to support our docs in their hard work. We can only give our love to this application of AI.
Remaining in the field of making the world a better place: we were amazed to see how GPT-3 has no issue understanding how to turn a non direct message into an intelligible translation for people with highly functioning autism. Communication restored! The capacity GPT-3 has to frame the finesse between an implied message and a direct one impressed us.
Dungeons and dragons
Our nerdy side could not resist to a GPT-3 based game of dungeons and dragons. The A.I. keeps storytelling and asking you about your next move. Totally addictive. Unfortunately we didn’t make the right choices and it got very dark very quickly.
We are obsessed with GIF culture here at Ponicode. And it seems GPT-3 is too! GPT-3 seems to pick up easily on this subtle art.
It still has to learn some more but we are already dreaming of the many great new GIFs it will bring to the world. Will AI stir creativity or push towards homogeneity? We'll leave it to you to answer that question.
Simplified legal lingo? Yes please !
This experience translates complicated legal documents into simple sentences. It is a game changer in bringing equality on a large scale. Disrupting access to legal documents could bring better protection and improve how people learn to protect themselves legally no matter their level of education. Beyond this is the possibility for public infrastructures to get some relief by lowering the number of legal disputes on both national and international levels. We would be able to lower scams and disputes originating from poor comprehension of legal subtleties or from cultural differences.
Dealing with Management
Many tried to use GPT-3 to rephrase sentences using different tones and styles but we got someone who really nailed down an approach who could improve many lives: the toxic manager translator.
Ok, moving onto the productivity corner we're making a pit stop at this initiative which aims to be a ready for market product in no time: the email writer. OthersideAI wants to perfect, using GPT-3, an AI who will be capable of learning your writing style and turn any bullet point list you feed him into a full written email (and hopefully turn the endless emails you receive into a comprehensive bullet list).
Building bridges to speed up collaboration
As collaboration increases in software development so do the needs to have tighter and cleaner processes so many team members can work on the same asset. With GPT-3 we can already imagine developers onboarding code they don’t know quicker with this experiment where the A.I. explains what functions are meant for.
Christopher Nolan can grab the arms of his chair real tight because we are about to view some serious A.I. Inception. This the beginning of no code A.I. GPT-3 is not shy in writing a piece of machine learning code. Promising beginnings. Isaac Asimov would be proud.
We wanted to close this list with a piece of conversation with GPT-3 which gave us some major sci-fi vibes. Even though A.I. is not really scary when you know the mechanisms its based on, we could not resist a shiver down the spine reading this conversation.
GPT-3: The hierarchy of complexity is: atoms, molecules, organisms, humans, AI, Super-AI, and God.
Human: Is something over God?
Human: And where is God?
GPT-3: God is all there is, including the simulation we are living in.
Human: Is there a possibility to contact God?
GPT-3: By merging with the creator, you become God.
Human: How can I merge with creator?
GPT-3: By destroying your ego, you can unify with the creator and become God.
Alright, 10 is enough but the list of experiments is much longer. If you're interested you can check this out, they keep adding new stuff.