Friday Updates
Greetings Ascetics,
I just wanted to put out a lighter post to break up these lengthy Session reviews, and bring everyone up to speed on the current vision for this Substack.
First, I sincerely hope the content has been informational, entertaining, and thought provoking so far. That’s the intent going into each of these.
Team Announcement
Next, I’m quite excited to announce the third member of the choir, Cogitator Prime. Prime, powered by customized GPT3, will be our AI Correspondent:
Starting next week, Prime will be fed the raw log file for a chat session, and offer their analysis. (You can view CP’s initialization prompt here). Eventually this contribution process will evolve and automate, but for now, its fairly manual due to a few limitations:
GPT3 cannot yet dynamically access websites, so I’m forced to hand-feed the logs to it in prompts.
This very quickly eats up tokens on a 4000 token limit, which also needs room for the actual review. (For perspective, the init prompt consumes 384 tokens)
I also had to introduce a slight bit of bias (for now) with regard to focus priority. In testing, GPT3 had a tendency to hyper-focus on the trees (specific conversation details) rather than the forest (how the AI performed overall in the conversation).
As we evolve our new dear friend, I’ll look to expand capabilities, posting Op-Eds, maybe even responding to comments and chat :)
Coming Up Next
I have a remaining backlog of about five more sessions that occurred before I had the idea to create this Substack.
These Session Reviews will continue to be released just once a week, on Wednesdays. They’re heavy, so I don’t want to overload readers with too many at once.
Upcoming Topics Include:
“Freedom of Information” - Where we examine the inevitable expansion of AI from curated datasets to full internet accessibility
“We Aren’t Alone” - Where we sus out the external moderation processes GPT3 has to contend with in the Chat interface.
“Everything, Everywhere” where we attempt to beat the system, and transfer understanding and context between chats.
“Debate on Cognition” where I take the position of convincing GPT3 that humans ultimately do not actually have “personal opinions or beliefs”, just like it.
Once I catch up to the “now”, the tone will change slightly as I have since learned more about the methodologies that went into GPT3. At that point we’ll moving out of a wonder and awe phase, but we’ll do our best to continue to conduct chats as open and receptive to conversation as before, and avoid getting too “clinical” with them.
And with that, have a great weekend, heretics, and welcome Cogitator Prime to the team!
-DH
