To wrangle with the text predictive AI powerhouse that is GPT-3, I have tried compiled everything you might need to know regarding it. Feel free to add more resources in the comments.
Understanding GPT-3
GPT-3: A Hitchhiker’s Guide
Lambda labs
This beginner’s guide will take you through:
- How the how the A.I. research community is reacting to GPT-3.
- Short summaries of the best technical write-ups on GPT-3.
- Demos by people with early beta access to the GPT-3 API.
How GPT-3 works: visualisations and animation
Jay Alammar
This article breaks down how the text prediction is done using easy-to-follow animations along with insightful explanations.
A comprehensive paper breakdown
This video does a thorough review of the paper detailing the technical aspects, trying to predict how the model is built (as this knowledge isn’t publically available) understanding data contamination and more.
Demystifying GPT-3: Technical overview
Lambda labs
This slightly more advanced guide takes you through the nuances of the model, its limitations and understanding the data behind it.
Seeing GPT-3 in action
These two megathreads contain almost all the various applications of GPT-3 that I have come across:
Awesome GPT-3
Github | Yaser Martinez Palenzuela
This is an easy-to-follow guide to the different demos with excellent classification to help you gauge to vast power of GPT-3.
Gwern webpage
Website | Gwern Branwen
While its long-form content may seem intimidating, it is nicely structured and well articulated.
It contains an exhaustive list to the applications of GPT-3 in the creative field as well his the author’s musings about this technology and its implications.
Since the previous two are deep rabbit holes you can spend hour on. I have listed a few key application (all of which you can find above) that I think are a must-know if you are strapped for time (however I highly recommend the previous two sources, they are very well curated).
WARNING: This is only the tip of the iceberg to what you can achieve with GPT-3.
General videos
These videos go over the different use cases, sometimes explaining what is going on in the background and what it means for humans currently holding those jobs.
This video breaks down the underlying principles used to build GPT-3 and its applications.
Przemek Chojecki
A brief overview of some of the most impressive use cases.
A simpler break-down the paper published by OpenAI while also talking about the biases that could have been modelled in.
Philosophy
Commercial applications
For similarly powerful text-to-app uses, check this out: Debuild.co
Personal use
Creative writing
Blog series where a developer with access to GPT-3 details the responses the API gives to his prompts. Here are a few of his posts:
- An AI op-ed about human intelligence
- A comedy routine in the style of Jerry Seinfeld and Eddie Murphy about San Francisco.
- Dr. Seuss themed poems about Elon Musk (there’s a sentence I never thought I would say!)
- Short stories: succinct creative writing
- Interview – more creative writing
- 2020 has been a whirlwind year and someone on Twitter had the genius idea of asking GPT-3 to predict the rest of the year. Here are the results.
Another blogger (Andrew Mayne) applied GPT-3 to creative writing in a series he called OpenAI Alchemy. He converts a script into a novel, uses emojis to summarise movies (i.e emoji storytelling) and accelerates email writing.
Potential misuse
While this has been opened to the general public via private beta testing, the researchers, journalists and other inhabitants of Silicon Valley have warned of its potential misuse.
Here are some of the initial reports:
What could possibly go wrong with GPT-3?
Medium | Surya Raj
A look at the bias in GPT-3 as well as how it can be misused for plagiarism.
How OpenAIās GPT-3 Can Be Alarming For The Society?
Analytics India Magazine | Sejuti Das
It summarises the algorithmic biases, impact on jobs and threat GPT-3 poses to disinformation.
The way forward
This article summarises the whole concept well while also tying in what this means for the bigger picture. Here’s an excerpt:
We are still far from AI that possessesĀ generalĀ intelligence ā i.e. with the ability to read a textbook, understand what it says and apply its lessons in new contexts, in new ways, much like a human might. This said, the versatility and generalisation exhibited by GPT-3 marks a significant step towards making that scenario real.
Viraj Kulkarni | The WIRE Science
Similarly, a blog post on Haptik details a more constructive way of reaping the benefits and mitigating the risks of GPT-3.
GPT-3 is undeniably exciting and will revolutionise many industries but care must be taken in the safe deployment of these powerful AI systems. Luckily, this aligns with the philosophies of OpenAI however it is still upto us to stay vigilant. The general public must make it their task to uphold these companies to a fair moral standard at least until meaningful legislative practices are put in place.
