A lot of this week was dedicated to watching people do stuff, honestly.
First I'll give an update on the game. Apparently we ended up getting an actual development team for Unreal, so we don't have to piddle around with stuff we don't know, so hopefully it comes around. We also secured some funding which is always good news.
For work, here's a play by play for what I did each day:
Monday, my boss wanted everyone to send me some example human responses rather than the AI responses that we'd been working with till now. He wanted to do a little comparison test between the AI and what actual human responses would be, so we set up a "reverse Turing test," in which I'd give him and another person a file with a 50/50 chance of either being all AI responses, or AI responses mixed with human responses.
It did seem fairly easy to tell what was AI and what was human based off what he told me the following day, but they supposed the AI responses were satisfactory.
Then the next day, my boss wanted us to do a 4-person task to implement a sort of flow between the front end and the back end. However, somehow all four of us misunderstood what we were supposed to do and thought we just needed to brainstorm a bunch of introductory messages for the bot. So we did that, but the next day we had to answer for why we didn't accomplish what we were tasked to do.
Thankfully, it was an opportunity to sort out further issues with communication. Hopefully something like that doesn't happen again.
After that little debacle, my coworker and I rode along with our boss as he set up the Azure server, along with a virtual machine running Kasm. Very cool tool, by the way. I hope we can use it in the future more because it would make working on the go a lot easier.
We learned a lot about how to set up the Azure environment and how to access the remote machines, and what to look out for when we are trying to create new virtual machines. There's a lot to learn there and even the 3 hours we spent o that wasn't nearly enough.
He also shared with us an interesting Python notebook on distilling the GPT-4 knowledge base into GPT-3.5, since OpenAI has fine-tuning abilities now for businesses, which can find below:
https://github.com/sixvo-labs/Sapir.git
Thursday, I was tasked with creating some MermaidJS diagrams detailing possible user stories when chatting with our chatbot. I came up with a bunch of different scenarios, which should cover a lot of our bases, but there are always going to be edge cases to think of. Some of the funnier ones were where the user was trying to troll the bot, and another where the user was getting frustrated at the smallest thing.
Today, I was tasked with organizing all the information I found from generating that data, identifying all the "actions" that the bot could take, such as "analyze," "match," etc. and put it all in a table with its different functionalities and scenarios. It's a pretty big ask but it should be valuable to do so.
Anyway, that's my account this week. Cheers.
Top comments (0)