Estimating OpenAI's GPT-3 Access/Beta Invitation Rates (Part II)
In January, I wrote a piece where I estimated the minimum number of people who now have access to GPT-3, based on the growth of the GPT-3 slack group.
Here’s what I wrote:
When I first joined the group [roughly September 2020], it was at about 2.2K members. All members in the #general chat had received access to GPT-3. However, when I checked the group today [January 11, 2020], it is now at 16,145 members.
Checking today, it is now at 44,960 members. This represents an approximate 2.78X increase in GPT-3 access, which is a pretty impressive scale-up technology, safety, and community wise.
Again, it’s worth noting that the membership count in the OpenAI beta community Slack group is the floor of the total number of people who have access to GPT-3, simply because many people who have access don’t join the Slack community. But I still think it provides us some indication of access, growth , and developer acceptance rates.
OpenAI recently announced that they are working on improving their toxicity filter, which I can imagine is one of the main challenges they are dealing with now safety-wise; preventing them from opening up access more. I can only speculate, but as we start seeing more rapid scale ups combined with reductions in overall GPT-3 toxicity, we may begin to see OpenAI approach offering complete public access to GPT-3, which is an exciting thing to think about.
Update:
Ryan Sevey has shared some interesting, publicly available traffic stats with me as an additional measure for estimating OpenAI’s acceptance/growth rates. The OpenAI beta saw 350,000 website visits in February of this year:
https://www.similarweb.com/website/beta.openai.com/