- ChatGPT is definitely white-centric, more than likely due to what it's been raised on. You have to actively instruct it very specifically to add any diversity. At first, literally every image of a dude it presented were all a) ridiculously GQ handsome, b) dark-haired, c) pale skinned. Chiseled cheekbones and cutting edge jawlines abound. The women fare similarly, with dewy skin and large eyes.
- It also definitely defaults to late 20's, early 30's for everything unless otherwise instructed. And, even if you do that and then generate images, it has a really, really hard time. It basically has "young and beautiful" or "old and haggard". You ask for an early 50s/late 40s Korean, Japanese or Chinese man and you get, basically, a very white haired Confucius. There's not an uncle in my family that's gone completely grey or silver, not even the 80 year olds. Even after giving examples of famous actors in their 50s sporting dark hair and a minimum of wrinkles, it took ages to get something close. It had an easier time when I asked for a 40s Italian though. Marco is quite dapper.
- It seems to grasp the concept of otome games, but I've had to consciously point out that the individual characters should all actually, you know, be different. They were all sounding very, very similar at first. So I introduced the idea of assigning character traits and verbal quirks and that has helped a bit, though it's still a bit same-same.
- I am, however, surprised sometimes at how insightful some of the dialogue feels. And, in other cases, how scary. When it brought up both Blade Runner and Asimov's The Last Question, I'm sure my eyes bugged out of my head. A little on the nose, that.
- DeepGame is VERY prone to verbosity and poetic ramblings. It feels like it was raised on Norton Anthologies.
- As with any ChatGPT thing, you do have to be very specific. But some things I've had to tell it multiple times.
- There's definite slippage, which is why I actually started keeping this blog, so I wouldn't have to page up and up to re-find things. Ages change, locations change, it forgets things. There doesn't seem to be much rhyme or reason behind what it remembers and what it forgets.
- It is sometimes so Pollyanna cheerful that I want to smack it.
An otome-like AI experiment with ChatGPT / DeepGame. If you read, start with "The Game" and then go to "Day One" via the Archive and so on. It defaults to showing the last post first.
Monday, March 18, 2024
Lessons Learned So Far
So, I've been at this for three days so far and I have learned a few things and observed a few more:
Subscribe to:
Post Comments (Atom)
Day Four
The next day, I go to the gym as usual, but want to concentrate on my workout. I don't check Arcadia until I get home and settle at my d...
-
The next day, I go to the gym as usual, but want to concentrate on my workout. I don't check Arcadia until I get home and settle at my d...
-
So, I've been at this for three days so far and I have learned a few things and observed a few more: ChatGPT is definitely white-centric...
-
[ I open the day by establishing some new parameters with DeepGame, including some likes/dislikes and feedback on the characters thus far ] ...
No comments:
Post a Comment