-
3 weeks of backlogged updates!
Since the last update, our not-so-friendly talking head project has undergone some heady 13th-hour changes–primarily focused on dropping Convai entirely in favor of the previously aforementioned ChatGPT approach used by Nathan. Switching Systems We had been holding out hope with using Convai despite the frankly uphill workability of it because its creator tools were so…
-
Pores and prescriptive emotions
Where previous updates had me puzzling over Convai’s esoteric State of Mind wheel, talking with one of Convai’s developers revealed that particular feature is still in development and basically, isn’t functioning consistently at the moment. So while that’s discouraging, it’s not a complete deal-breaker for Convai. Its capabilities to deliver good responses and character-building tools…
-
Convai updates & tweaks
Last week, I experimented with making a less helpful character using Convai, and this week I wanted to keep tweaking with that personality and see if I can limit even further the information the character would be willing to give. Before that, though, it seems the Convai developers have updated their webpage playground for testing…
-
Making a rude boy
Last week I focused on a character I’d been writing for years. This week’s experimenting with Convai was more word-salad-focused making a character out of thin air. One of the problems encountered last week was the absolute cheery helpfulness of the character. She was willing to divulge every single detail asked about her, even topics…
-
Back for part 2
In starting the second round of New Media Project, now with 3x the amount of teamwork. The plan over the last three weeks has been the development of automated avatars and NPCs–that is, using generative text and voice-to-text tools and bring digital characters to “life”. The whiplash is not lost on me, considering my previous…
-
A sliver of progress!!!
(written Mar. 17th) After last week’s diversion, I returned to my attempts to use my own training data on simple Google Colabs. One of the Shakespeare generators, found on a Tensorflow tutorial, has finally worked enough to let me train a model on my own data! Sort of! It still looks like absolute gibberish, depending…
-
A last-minute diversion (p.2)
(written March 8th, 2023) Anyway, Novel AI’s writing tools. The generation and customization options here are bountiful, very smooth, and fairly straightforward to understand. That being said, aside from some sections where I can once again enter my writing to keep the characterizations on track, there is nowhere for me to individually train a model…
-
A last-minute diversion (p.1)
(written March 7th, 2023) I haven’t finished experimenting with Ruslan Brilenkov’s tutorial yet, as it’s estimated to take roughly 3 hours to run just one training model with his method, but I have diverted briefly to trying other generative programs at the recommendations of my instructor, just for the sake of having something to present…
-
A last-ditch effort
After taking a week to recenter from. Well. Lots of failure. I’m still looking for ways to train a simple program. I’ve stepped away from Colabs for now (because none of them have been working; they’re all either out of date or just. Don’t seem to function the way their original authors thought they would)…
-
I partially gave up this week
I’m reaching a point of frustration that I don’t know how to dig myself out of. Every tutorial I’m finding online about training your own narrative model assumes the person looking has some semblance of experience with Python or other coding languages, or the supposed plug-and-play methods like from Google Colabs are outdated with their…