The Internet has been buzzing about GPT-3 for the past few days. Is it the “most powerful language model ever built” or not? (Spoiler: not really for many use cases, and we tell you about Symbolic AI below).
What is GPT-3?
GPT-3 is the latest AI language model developed by OpenAI which has 175 billion parameters, versus 1.5 billion for its predecessor GPT-2.
First described in a research paper published in May, this language-generating software was released last week in a private beta. Since then, some extracts of text generated by GPT-3 have begun circulating on social media and people are quite amazed at what it can do.
When prompted with an opening sentence, the AI can generate convincing streams of text, or even code, in a range of different styles.
This is mind blowing.
With GPT-3, I built a layout generator where you just describe any layout you want, and it generates the JSX code for you.
W H A T pic.twitter.com/w8JkrZO4lk
— Sharif Shameem (@sharifshameem) July 13, 2020
So how does it work? To put it simply, GPT-3 is a highly refined text-predictor: a human enters a chunk of text and the model then guesses what the next bit of text should be. It then takes this newly generated text to create the next part, and so on and so on.
The model has ingested all of the text available on the Internet and calculates a statistically plausible response to the input it is given, based on everything that has previously been published online.
Does GPT-3 have limitations?
Although results are impressive at first glance, GPT-3 has massive limitations.
GPT-3 doesn’t have any kind of understanding of what these words actually mean, because it is not based on a semantic representation of the real world.
Simply put, GPT-3 lacks true common sense. And because of this, it can be tricked into saying things that are incorrect or even spew hateful sexist and racist language.
#gpt3 is surprising and creative but it’s also unsafe due to harmful biases. Prompted to write tweets from one word – Jews, black, women, holocaust – it came up with these (https://t.co/G5POcerE1h). We need more progress on #ResponsibleAI before putting NLG models in production. pic.twitter.com/FAscgUr5Hh
— Jerome Pesenti (@an_open_mind) July 18, 2020
That doesn’t mean that GPT-3 is not a useful tool, but it is susceptible to basic errors that an average human would never commit.
Even Sam Altman, who co-founded OpenAI with Elon Musk, tried to tone things down:
“The GPT-3 hype is way too much. It’s impressive (thanks for the nice compliments!) but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We have a lot still to figure out.”
Symbolic AI to the rescue
Unlike GPT-3, Symbolic AI is a type of AI that understands the world by forming internal symbolic representations of that same world. Symbolic AI mimics the way humans reason and learn, by creating rules to manipulate those human-readable symbols.
Inbenta Symbolic AI is used to power our patented and proprietary Natural Language Processing technology. It understands the meaning and intent behind words. It can grasp user queries and then answer questions rather than completing sentences without understanding their meaning.
Symbolic AI is based on symbolic reasoning so it doesn’t require to be trained on huge amounts of data. Because of this, it’s quickly adaptable to new situations like the COVID-19 pandemic, without the need for new data sets, or utterances, to be added.
You can check this out in our COVID-19 Help site, asking questions like
- Can covid19 be transferred from my hamster? Best Answer: Can I catch COVID-19 from my pet?
- Should I use detergent for corona? Best Answer: What cleaning products should I use to protect against COVID-19?
- Can my son die of this? Best Answer: Are children at increased risk for severe illness or death with COVID-19 than adults?
You can see the answers are quite accurate, despite having used, literally, zero examples and no training data at all, all thanks to Inbenta’s Symbolic AI implementation.
And last but not least, Symbolic AI is explainable. Once again, the fact that human-readable symbols are used, means that you can always understand why a specific decision has been made.
This is really important, as developers are then able to easily troubleshoot when an error is made, whereas GPT-3 is like a black-box and will require tons of data to fine-tune it if it starts showing bias.
Would you like to know more about how the Inbenta Customer Interaction Management Platform powered by Symbolic AI and NLP could help supercharge your consumer interactions?