Is the data pool only as of a certain date? Or is it supposed to be dynamic/current?
(Sorry if regular users aren't supposed to comment here, I have no idea how these kinds of websites work, feel free to ignore me or delete this if I'm out of line... 😅)
I tried searching for a super popular video game side-character that emerged in September of 2021, but this AI has absolutely no idea about him (with hilariously unrelated yet consistent results depending on accompanying terms, LOL). So, I take it the system is drawing from a static data set, perhaps from the "Model Card" section's original version's February 2021?
Is that by design, or did I accidentally find a blind-spot? I might as well also ask if that's planned to change at any point in the future? Thanks.
I would imagine it would have something to do with copyright, game devs are notorious for throwing around takedown notices for the slightest infringement - even fair use has a bad day of it too, it's kinda gross imo but what can you do? 🤷 🤣
I would imagine it would have something to do with copyright, game devs are notorious for throwing around takedown notices for the slightest infringement - even fair use has a bad day of it too, it's kinda gross imo but what can you do? 🤷 🤣
Oh, that's definitely NOT the case for Toby Fox (indie dev) & his games Undertale & Deltarune. He's VERY open about letting everyone do just about everything with his franchises. The AI uses data from both Undertale (released in 2015) and Deltarune: Chapter 1 (released in 2018) perfectly fine. Chapter 2, meanwhile, came out in September 2021, & took off with even higher popularity than Chapter 1, especially the character named Spamton (a creepy crazy personified spambot/puppet).
But yeah, trying to get craiyon to create images with "Spamton" results in like, quaint European villages with thatched roof cottages, "Spamton G. Spamton" generates black & white photos of distinguished gentlemen with 1800s-era facial hair, & "Spamton Deltarune" makes unholy amalgamates mixing several Deltarune Chapter 1 characters together. 😅
So, it looks to me like the AI is limiting its data pool's date range via intentional parameters, or it's a stationary offline pool that isn't being updated. Or the Google TPU is only allowing access up to a certain date? Or the devs accidentally have the range set to end at the particular date they set it up instead of a fluid "today," or somethiiiing? I dunno, just throwing darts blindfolded here. But as popular as Spamton is now, with TONS of fan content created daily, Top-3 trending on Twitter several times a month every month, & about 9 months having elapsed, if it was using current & relevant pools, it would definitely know him by now. 😉
It'd be nice if that could be expanded to keep up with trends as they develop (like new meme formats constantly being born)... Though of course, this is likely a very complex project, & restricting the data pool might be more helpful to fine-tune & stabilize results before letting it venture forth into the wild. So if that's on purpose, I get it.
Still, hopefully someday it can be set free (ahh, character relevance... 🥲 ). ...Or
if it's SUPPOSED to be pulling from a live pool & there is, in fact, a bug or error preventing that, then I hope it can be remedied soon. 😅
(Edited to more clearly word my guesses of why this might be happening. Now it's been 11 months since DR Ch 2's release. If anyone knows the answer to what the case is with the data, please reply whenever you have the time to. I'm curious in general, & it might be handy for us laymen users to understand why we might not be able to generate something based on recent events or media [knowing what the cutoff date is would help a lot]. And whether or not that's expected to change in the future. 🙂)