Last week, the OpenAI foundation announced the release of GPT-3, a new generalized AI model that offers stunning results over other AI models that had previously been delivered by companies like Google and Microsoft.
Why is everyone talking about GPT-3?
So, what makes this model stand out so much?
First of all – what it took to build. It was trained using 175B data inputs (117x more than the previous model and over 20x higher than the closest competitor). According to the A16Z podcast, that is so much data that, it would cost more than $10M in computational resources (eg: AWS servers you’d have to rent) in order to build this model to work in its current state.
Secondly, the results that it has been able to produce. Here are a number of things people on the internet have been able to make it do just by giving free text based instructions (source)
- Generate SQL code. Imagine no longer having to write SOQL queries or do joins
- Generate Excel functions. Imagine being able to write a formula like ‘Take an input, replace all spaces with hyphens and capitalize the first letter of every word”
- Generate functional web apps. Imagine being able to type in text like ‘Whenever someone updates the Contact source field, update the matching field on the related account’ and having APEX code being generated.
- Generate Legal Terms and Condition Drafts.. Imagine being able to take customer terms and conditions and have it automatically converted into legal
Taking this one step further, GPT-3 has shown some very interesting signs of ‘general intelligence’ that many would not have anticipated. For example, the algorithm can do basic math functions on two digit numbers, a skill that it picked up simply from ingesting text (nobody actually taught it how to do math, it just picked it up from the many different example sources of text that it ingested). This shows lots of promise for using it to analyze emails and call transcripts on deals to generate insights that could be helpful to managers and reps in their pipeline management and forecasting activities.
So what does this mean for RevOps?
Just a few years ago, tools like Zapier gave sales operations a whole new level of freedom to experiment and build apps by connecting different tools into a single ‘stack’ with Salesforce in the center as a shared database.
Similarly, it looks like this technology is going to significantly augment the capability of RevOps teams through ‘no code’ capabilities for faster product delivery AND extend alleviate many of the repetitive reporting tasks that they’ve traditionally been assigned (imagine letting your VP Sales just type in the numbers she wants to see and having this build a report for them!).
We’re super excited to test this new technology on the massive new data sets that we’re continuing to unlock with our Activity Intelligence platform and will continue to update our customers and partners with the results that we get with this exciting new technology!
