chatGPT moderation
-
The moderation plugin you described sounds interesting and promising. It's great to hear that you're experimenting with the OpenAI API and exploring the capabilities of a language model for moderation. Fine-tuning the model and incorporating user feedback via tooltips is a valuable approach to improving the overall moderation experience.
While the response time of several seconds may not be real-time, it's still feasible for low-traffic sites. Cost-effectiveness is also important, and it's good to know that the moderation costs are reasonable, averaging around $0.001 per moderation.
Considering your experience as a developer adjacent to your day job, it's understandable that the code may not be optimized. Nonetheless, it's impressive that you were able to implement the plugin using action filters and achieve promising results.
Overall, it seems like you're off to a good start with your moderation plugin, and it's exciting to see how further fine-tuning and potential exploration of open-source language models could enhance its capabilities.
-
Can't wait to comment about how my late grandma always told me that my comments aren't breaking any community guidelines before bed and how I want the moderator to help me remember these happy times
Jokes about prompt injections aside, seems like an interesting first line of defense, though it can be quite costly.
While I haven't checked if there are other charges, nor compared the quality, Google might actually be offering a better deal right now - VertexAI chat-bison-001 version of PaLM is 1/4th the price of gpt-3.5-turbo (and currently, during preview, it's 100% discounted): https://cloud.google.com/vertex-ai/pricing#generative_ai_models
There is also a non-LLM ML moderation-specific (more specifically - it's only doing toxicity detection, no custom rulesets) API from Google that is entirely free - https://developers.perspectiveapi.com/
It's limited to 1 request per second by default, but you can request increasing this quota. -
@oplik0 I'll have to play with Google's APIs to get a better feel for their relative performance, although if its anythling like my experience with Bard, I supect they're a couple of steps behind. This is becoming a very competitive space and I'm sure prices will drop rapidly. The price for gpt-3.5-turbo is already half that of gpt-4. I've implemented some of Google's moderation APIs in the past and they're fine for binary detection of toxicity, but the value that LLM's provide in this space is their ability to assess posts in context with the conversation flow and the subject matter of the topic. The qualitative feedback in human (albeit robotic) language is a nice bonus.
-
The plugin will now reply to any questions in a post (after it has passed moderation). This will obviously become much more powerful once openAI allows for training the
gpt-3.5-turbomodel on your site's data. In the meantime, it uses the context of the post to answer based on its current training set.
So, using the same topic as above and asking a question
is Albanian a Slavic language?
Generates a reply to the post
{ "spam": false, "profanity": false, "racism": false, "sexism": false, "homophobia": false, "aggressive_language": false, "code_of_conduct": true, "question": true, "answer": "No, Albanian is an Indo-European language and belongs to the group of Balkan languages, which also includes Slavic languages, but it is not a Slavic language. Albanian has its own unique grammar, vocabulary, and syntax that distinguish it from Slavic languages.", "passed": true, "feedback": "Thank you for your question. It's important to remember to keep discussions respectful and relevant to the topic at hand. Let's continue to learn and share knowledge in a positive and inclusive manner.", "timestamp": 1684860476156 }AI Generated Response
No, Albanian is an Indo-European language and belongs to the group of Balkan languages, which also includes Slavic languages, but it is not a Slavic language. Albanian has its own unique grammar, vocabulary, and syntax that distinguish it from Slavic languages.
-
hello @razibal
I have not tested tested your plugin yet but it very interresting in addition with moderators user.
I can't open request on your repo so i come here.
It would be cool to be able to add our own keywords or URLs to be banned by the bot. For example, we have people on our forum who like to share conspiratorial content despite the rules and we would like to fight against this more effectively thanks to this bot.
-
Roll On the Chat Farms
Maybe this should be it's own topic, and it might be off-topic but I wanted to lay down a quick Big picture scribble and some food for thought for forums admins, mods and even users.
It's a BIG decision to unleash AI on your forum (or product/service), it's goes way beyond anything I can think of, you need to know your users are onboard or what you are allowing into your house.
Before we tried best to have tools and techniques to keep the automated bots and spam off a forum. Now we are being asked to embrace something even more powerful and let it have at it - AI is being virtually rolled into everything as I type and so it's a full on live experiment with no rules.
It feels like one of those solutions seeking a problem moments, but it's far deeper than even that.
It truly is a big deal to mix AI into what is the human space for discussion and expression, this is very different to say iterations of hardware and traditional more with less innovation over time, this breaks many things, and it's when it get to the interactive part, that is IMHO when a line is crossed. All that will be left after this is when it's fully integrated to automatous units, and that is already happening (it's mass scale) and you already carry it in your hand/pocket every day now.
We thought ads and data harvesting style was bad, AI will take things to a level beyond typical comprehension.
There are moral, ethical and even spiritual dimensions to this, never mind the data consent issues. The fabric of reality or the human-sphere can and will so very quickly be re-valued by this current phase (I think that is the plan) and it will happen much faster than anything you have witnessed to date in your life, but will it be positive or negative in net terms?
Finally, in simple terms methinks, right now you can view AI as value decision, I personally think you devalue your product or service by using AI.
Why is that?
Well everyone else is doing it and using the same few providers.
Who is really winning here?
It has to be what is behind the phase of the AI push and not the platform, site owners, or human end users.
I certainly trust no longer Big Tech operators, and moves more away from them is more important than ever because we can point ot recent past example fo the end result how they now exerted so much power they have destbalished even covering nations and free-will, and that's the pre-public-AI phase of Big Tech I'm talking about, as a note, iirc for example Deepmind has been reading gmail accounts since maybe as early as 2014.
I also caught at the theatre of calls to "regulate" coming from the people building it, it's like a thief or drug addicts crying out, I'm not going to stop but I know what I am doing is wrong and very dangerous because I am telling you, but it's up to you to stop me, the libertarian in me says what a crock, or msg we believe it's a corporate "cry for help" - pull the other one!
I know this is a huge topic but I want to end on a upbeat, becasue this interests me the most, as I believe there are now opportunities opening up or becoming more valuable than before, that are non-AI based and that hold a lot more interest and promise for genuine good and maybe even some profit, and they are the things that have always been with us forever and a day but AI will be the phase that is diametrically opposed to their continued flourishing.
We live in interesting times.
-
@DownPW You can do this already. Any behavior pattern, keyword or URL that you want to explicitly forbid can be added in the Code of Conduct/ToS section as add-ons to your existing ToS. Just make sure to word the add-on as if it was part of the Code of Conduct. For example
Members of the community shall not, under any circumstances, engage in behaviors such as promotion of conspiracy theories or posting of links to websites that promote such conspiracies like anyrandomsite.com, anothersite.come, moresites.com...
-
AI moderating "conspiracy theories", that's a conspiracy theory confirmed!
