News for hackers. Stuff that matters.
Re:AI Dungeon will block certain words, review content flagg... by inopinatus on Wednesday April 28, 2021 @08:07AM (#26968398)
Good grief: "Latitude reviews content flagged by the model" - or, as it was put in another forum: every time the AI flags your game, a Latitude employee will be personally reading your private content.

The key reason is perhaps this, buried deep in the text: "We have also received feedback from OpenAI, which asked us to implement changes". Given the volume of prompts that AI Dungeon throws at GPT-3 in the course of a game, it's easy to conclude that Latitude has a real sweetheart deal on the usual pricing, and that they basically have to follow orders from their benefactors.

Whatever may be said of the robocensor they've thrown together - and early anecdotal reports are, it is painfully crude, both oversensitive and underspecific - how they've handled communicating the change is extraordinarily naive. Not for the first time, either: Latitude has form on suddenly imposing major service constraints in a peremptory, underhanded fashion that infuriates their customers. Repeating past PR mistakes, and now doubling down by complaining about "misinformation" and throwing shade onto others, is starting to look like a pattern.

Re:AI Dungeon will block certain words, review content flagg... by lainga on Wednesday April 28, 2021 @04:19PM (#26973672)
Thus far I have seen screenshots of it flagging the phrases "I would like to buy 4 watermelons" and "I just broke my 8 year old laptop". Regardless of your opinion on the ethics of this feature it seems to need a little polish.
Re:AI Dungeon will block certain words, review content flagg... by cjhveal on Wednesday April 28, 2021 @06:29PM (#26975060)
Reminds me of the 2005 conversational simulation game, Facade[0], in which any mention of the word "melon" would be met with being immediately kicked out of your host's dinner party.

[0]: https://www.playablstudios.com/facade

Re:AI Dungeon will block certain words, review content flagg... by inopinatus on Thursday April 29, 2021 @12:55AM (#26977611)
c.f also the “Scunthorpe problem”. https://en.wikipedia.org/wiki/Scunthorpe_problem
Re:AI Dungeon will block certain words, review content flagg... by josefx on Thursday April 29, 2021 @04:03AM (#26978533)
A working text content filter that doesn't randomly flag seemingly innocent input is something I wouldn't expect to see within my lifetime.
Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @05:30PM (#26974458)
Yeah, that would get flagged in human code review, too. Way too many magic numbers
Re:AI Dungeon will block certain words, review content flagg... by crooked-v on Wednesday April 28, 2021 @05:46PM (#26974647)
One number is 'too many'?
Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @06:01PM (#26974803)
I've gotten into way too many fights over this, and my bitter sarcasm is leaking here. I actually like numbers in code. But I've literally been told that

   distance = ((x1-x0)**2 + (y1-y0)**2)**.5
has "magic numbers" and that's a "code smell"
Re:AI Dungeon will block certain words, review content flagg... by ectopod on Wednesday April 28, 2021 @06:22PM (#26974987)
Raising to the power of .5 rather than using the sqrt function does whiff a bit. Even better, use hypot.

https://developer.mozilla.org/en-US/docs/web/javascript/refe...

https://docs.python.org/3/library/math.html

Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @07:38PM (#26975604)
Yup, hypot is better on performance grounds, if you assume I'm using doubles. But is it as readable as my snippet?
Re:AI Dungeon will block certain words, review content flagg... by Noumenon72 on Wednesday April 28, 2021 @08:32PM (#26976000)
How could it not be? It has a name that says what it's doing, while yours is more punctuation than text. To "read" yours I have to just skim over it and mentally assign "according to the variable name, that hodgepodge means 'distance'... I hope".

I have a BS in math, so obviously I'm familiar with the math notation, but I code like an English major.

Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @08:47PM (#26976133)
If the function was named "hypotenuse" then it would be unambiguously more readable. But since the function is considerably older than the modern zealotry surrounding code readability, it's this weird 5-letter word that most people aren't familiar with. Contrast that to the Pythagorean theorem, which folks are exposed to in elementary/middle school.
Re:AI Dungeon will block certain words, review content flagg... by seoaeu on Wednesday April 28, 2021 @10:56PM (#26976930)
Yeah, not to mention tons of possible variations of that line that you might not notice at a glace but would completely change the output.
Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @11:12PM (#26977011)
Yeah, my bad for not radiation-hardening my code

https://github.com/mame/radiation-hardened-quine

Re:AI Dungeon will block certain words, review content flagg... by seoaeu on Thursday April 29, 2021 @11:35AM (#26982505)
I was thinking more in line with http://www.underhanded-c.org, but that was also an interesting read
Re:AI Dungeon will block certain words, review content flagg... by sdenton4 on Wednesday April 28, 2021 @08:52PM (#26976173)
Well, clearly we need a metricFactory to provide the distance function, in case we later want our game to run in a non euclidean context.
Re:AI Dungeon will block certain words, review content flagg... by klyrs on Wednesday April 28, 2021 @10:01PM (#26976637)
I hope you get that metricFactory out of a computableMathConceptFactoryFactory
Re:AI Dungeon will block certain words, review content flagg... by echelon on Wednesday April 28, 2021 @11:38PM (#26977166)
I was once told in code review that using XOR was "too low level" for Ruby code.

This was by a person that sprinkled lazy backfilling behavior throughout the codebase, created magic method_missing dispatches against polymorphic ActiveRecord relationships, and exposed the database as a service through a RESTful DSL that allowed arbitrary method dispatch.

I'm glad I spend most of my time in statically typed, math and binary-friendly codebases now.

Re:AI Dungeon will block certain words, review content flagg... by fpgaminer on Wednesday April 28, 2021 @07:29PM (#26975536)
I'm a moderate AI Dungeon user, including some NSFW content.

Frankly, I've always assumed the devs have access to my sessions. Whether it's AI Dungeon themselves or OpenAI, you know that data is being harvested. And at this juncture in text AI development, that kinda makes sense. Obviously I would prefer privacy where possible, but these companies are data hungry and they own the park we're playing in. So it only seems fair.

We'll have to wait awhile for GPT-3 like models to democratize before we can expect real privacy. In the meantime, just err on the side of assuming all input to OpenAI systems is being harvested.

> and early anecdotal reports are, it is painfully crude, both oversensitive and underspecific

Funnily enough, I didn't run into any issues with any of my NSFW sessions since they implemented the filters. So I guess the problems are with SFW sessions so far :P

Anyway, I've got to say I'm kinda happy about AI Dungeon tackling this problem. They made it clear in their announcement that they aren't targeting NSFW content in general, just the one subject. The AI has a tendency of shoving that subject randomly into sessions, which isn't great. If they can eventually filter that filth out without affecting quality otherwise, I think the service will be better for it.

Re:AI Dungeon will block certain words, review content flagg... by minimaxir on Wednesday April 28, 2021 @04:42PM (#26973942)
It should also be noted that OpenAI’s content filters are extremely sensitive which may explain some of the downstream effects.
Re:AI Dungeon will block certain words, review content flagg... by spullara on Wednesday April 28, 2021 @04:18PM (#26973660)
They are implementing the same things everyone has to implement to be a public implementation. It is part of their terms of service. I went through the same process making a Slackbot that could sort of pretend to be me and other folks given different prompts.