Just canceled my subscription to AI Dungeon.

[kane_magus]

Apparently, they're "testing" the implementation of some kind of draconian filter to censor out all the naughty/violent stuff, and it has broken basically everything. And they did this without telling anyone they were going to be doing it. With this, on top of the energy bar shit and whatnot, it's basically not worth it anymore.

Ah well, it always felt like AI Dungeon was too good to last, and I guess it ended up being true. I'm just going to bail out now, rather than stick around and watch this thing die a lingering death. *weary sigh*

If they reverse course on this, I may come back at some point. Might even resubscribe, but as it stands right now, the thing is entirely unusable.

(EDIT) They have released a statement about this, finally.

Still waiting on my subscription to finish "cancelling," after which I'll be deleting my account entirely, removing the bookmark to the website from my browser, and washing my hands of this fiasco. (/EDIT)

(EDIT 2)

AI Dungeon and its userbase right now:

Watch on YouTube

(/EDIT 2)

Comments

[owsf2000]

Yeah, probably best you get out now while you still have good memories of the system I suppose. Reading over their little press release does raise some questions about how they run their platform.

1. "Continuously improve the AI based on feedback to be both fun and welcoming" - This isn't any kind of change I would hope. I mean, I assumed they were always doing this anyway... Almost like they added this to start the list just to make it look longer.

2. "Providing a way for users to report false positives so that we can limit the impact on other types of content." - You'd think they'd have this in place already, unless they simply weren't doing any kind of content policing at all prior to the big stink created.

3. "Inform moderators in advance when platform changes are being implemented" - This seriously shows they just aren't a very responsible group period. Should be basic common sense to inform at LEAST your own staff when you're going to release changes. They have to know what's going on if they're going to be able to explain to users what's going on.

I never did start an account in AI Dungeon myself. I didn't know it was a thing really (or don't remember knowing it was a thing) until you mentioned they were starting to rush towards a monthly subscription model. At that point I figured it was already too late to bother as I expected they would just get more and more expensive over time, and deteriorate the quality of any freebie offers to help force subscriptions. Which they were already doing in the initial plan if memory serves.

Their answers also seem to show lack of consistency which further shows to me that they either don't know what they're doing or they're trying to play both sides of the field.

"We are in the process of implementing technical safeguards and policies to support our community guidelines prohibiting sexual content involving minors in AI Dungeon. Additionally, we are updating our community guidelines and policies to clarify prohibited types of user activity."

In this case for instance, they've already implemented safeguards and policies to support community guidelines... but they're also under the process of updating their community guidelines. Reads like they added controls to forbid particular content that they hadn't made forbidden. The real problem being that they rushed it out without testing and necessary safeguards (flagging false positives, etc) and didn't warn anyone so the natural result happened.

I love how they try to silence the masses though.

"How will it affect my gameplay?"

"For the vast majority of players, it shouldn’t. It will only affect your gameplay if you pursue these kinds of inappropriate gameplay experiences."

ie: they're saying if you're complaining about it it's because you're pursuing that one type of content they're blocking right now. But as users like yourself have shown it's throwing up a LOT of false positives and breaking the AI. All without any way to report false positives, if they're in the process of adding that functionality.

"We built an automated system that detects inappropriate content. Latitude reviews content flagged by the model for the purposes of improving the model, to enforce our policies, and to comply with law."

That's the longest spelling of "Yes." I've ever seen to the question of "Are you reading my content." ;)

Hm. This is more of a rant than I really expected to have on a system I don't use. I think their press release nipped at my pet peeve regarding non-apologies from corporations. Note that not once did they apologize for fucking everything up. Their system, certainly, but they're charging money at this point and they wasted a LOT of peoples time over the system breaking due to what seems to be a rushed, half-hearted, and very untested update.

[kane_magus]

For what it's worth, if you wanted to just mess around with it for a bit, it is (at least for now) possible to do so, for free without making an account. It's just that most of the features (not the least of which being the more advanced "Dragon" model of the AI) is locked behind a paywall. The free version is cool for what it is (and even that has greatly improved from what it was when I first started using it... almost a year ago), but... yeah, at this point, I'm not sure it would be worthwhile to even get invested into it just time-wise, let alone money-wise.

The only censoring they had been doing before, that I was aware of, was replacing the word "rape" with "show respect to," which was pretty silly. Now, all of a sudden, the AI simply doesn't respond at all if whatever filter is triggered. It just throws the typical "The AI doesn't know what to say. Alter, undo, or try again" error message, which is what you got before when the AI had a hiccup or whatever.

And yeah, even though I never truly thought anything was actually private on there to start with (EDIT even though they had previously explicitly said that nobody at Latitude would be reading private stuff /EDIT), now that they've straight up come out and said they'll be reading your shit, "private" or not, yeah, that's the bridge too far for me.

In any case, I'm pretty much done with it. Already canceled subscription and deleted account. It was cool while it lasted, but they've pretty much cut off their nose to spite their face at this point. In their attempt to aggressively stop "pedophiles" or whatever (and I won't even get into the whole thing of "is it really pedophilia/bestiality/whatever if it's nothing but text on a computer screen?" on the one hand vs "hUrR yOu ArE aNtI-cEnSoRsHiP sO yOu MuSt Be PrO-pEdOpHiLiA hErPiTy DeRp" on the other), they've basically all but killed off their own project. It's not even a "freeze peach" issue, as too many of the critics are trying to claim, because Latitude is a private company and can do whatever the fuck they want, but it's still essentially the death knell for their product, all the same (outside of the few who will continue on/join up expressly because they're "pwning the pedos" or whatever *eye-roll*). Just because they can do whatever they want with their product, implement whatever heavy-handed content filters that they want or rifle though users' files whenever they want, that doesn't mean anyone is obligated to continue giving them money for said product. And like others have pointed out, not even multi-billion dollar company Google knows how to properly and consistently use AI algorithms to police shit like that without far too many false positives (*side-eyes Youtube*), so how can a tiny company like Latitude be expected to do it?

All that aside, hopefully, eventually, OpenAI will stop being such tightwads about how they parcel out usage of their shit (apparently there's a huge waiting list for anyone to even have a chance of using their tech) and what they charge for it (apparently it is exorbitantly expensive, which is why [ostensibly] AI Dungeon was "forced" to start using the FB-ized energy bars and shit). Honestly not sure why they call themselves "OpenAI" when it clearly isn't "open" at all, at least not by any understanding of the word "open" that I know. Some people were saying it was probably OpenAI who forced the hand of Latitude into censoring everything, based on OpenAI's own policies, but others have claimed that the Latitude devs said, on the Discord or whatever, that they (somehow) weren't bound by the strict OpenAI policies. If that's even true, then either the devs foolishly believed that was the case and then got fooled when OpenAI brought the hammer down, or they were lying about it from the get go, or else this actually does have nothing to do with OpenAI policy at all and really is just Latitude being excessively CYA. Doesn't really matter, ultimately. The damage is done, either way.

[kane_magus]

EleutherAI might be something to keep an eye on. (Though there's no "AI Dungeon"-like project attached to that yet, at least not that I'm aware of.)

[kane_magus]

For what it's worth, if you wanted to just mess around with it for a bit, it is (at least for now) possible to do so, for free without making an account.

Nope, never mind. Was just fucking around with it a bit, again, and you apparently only get a very small amount of actions before it cuts you off completely and prompts you to create an account now. Fuck that. Definitely not worth the time.