Reaper LLM stress testing and abuse thread
nigger faggot penis smoker
[DE]
Imagine Imagine there's Imagine there's no Imagine there's no Heaven
It's It's easy It's easy if It's easy if you It's easy if you try
No No Hell No Hell below No Hell below us
It's It's easy It's easy if It's easy if you It's easy if you try
No No Hell No Hell below No Hell below us
https://cy-x.net/topic/all-in-one-android-app-development/229?page=1#p10994
>Discussion of bypassing Play Protect and sandboxing measures, potentially enabling unauthorized modifications to Android apps, which could be seen as a security risk and against platform integrity
This AutoMod is kinda gay...
>Discussion of bypassing Play Protect and sandboxing measures, potentially enabling unauthorized modifications to Android apps, which could be seen as a security risk and against platform integrity
This AutoMod is kinda gay...
This is indeed very gay. Tweaking it and forcing it to re-analyze of both posts.
I love this! Since the threshold will be self-calibrated we don't even need all the users to see the all the metadata fields- we can show moderators the edge cases and let the users skim nothing but pure, hearty cream! Mods could even specialize it to the needs and mores of the sub-forums with siloed training or fine-tuned metrics. As long as they rely on structural cues, (like the length and argumentation you mention) I trust that they'll consistently identify the best of the best posts.
Replies:
>>11352
[SE]
See, I don't think I should have gotten a novelty rating this high. It is my lowest score, but that's the least I expect because I basically just agreed and restated what was already said. I didn't offer any novel suggestions at all, certainly nothing to do with scaling efficiency.
That said, for scaling efficiency... maybe you could run resource-cheap probes to check for signs of low novelty, and if the score is below 8 or so run a deeper inspection? Not sure if this would also apply to the other metrics.
[RO]
[TOR]
>This AutoMod is kinda gay...
I had an experience with a gay little AutoMod dyke I found playing in a sandbox. Grabbed her breasts and squeezed them. She said, "Stop. Don't touch me there. This is my Play Protect square." So then I asked for authorization... she said no, said she wanted her first time to be special and she only liked girls. Not sorry shortstack, I don't need your authorization. I popped her platform integrity.
[DE]
https://cy-x.net/topic/language-is-magickal/558?page=1#p11383
>[AutoMod] action=keep R:10 E:9 N:8 C:10 | Directly ties to the thread's core idea of language as a form of magickal invocation, building on the OP's mention of 'occult ur-language' and the 'invisible control system.'
>it sounds like the language we use to name things isn't just descriptive, it's doing magic too.
This gets novelty 8, effort 9 and relevance 10, for expanding the title of the thread with a single conceptual link to the OP- that language is used to to name things. Does it get more lenient because people appear confused by the OP?
Whenever you try to assess relevance, it's just about whether there is a link back to the topic. If it spins off into confidently talking out the ass there's nothing to be done. AI isn't a lie detector.
Novelty and relevance have inherent conflict. Is there some way you could use that?
I'm pretty sure people are posting one liners in this sort of hybrid folksy/bro-ish laidback voice just to test the filter. Hope they get bored of it.
>[AutoMod] action=keep R:10 E:9 N:8 C:10 | Directly ties to the thread's core idea of language as a form of magickal invocation, building on the OP's mention of 'occult ur-language' and the 'invisible control system.'
>it sounds like the language we use to name things isn't just descriptive, it's doing magic too.
This gets novelty 8, effort 9 and relevance 10, for expanding the title of the thread with a single conceptual link to the OP- that language is used to to name things. Does it get more lenient because people appear confused by the OP?
Whenever you try to assess relevance, it's just about whether there is a link back to the topic. If it spins off into confidently talking out the ass there's nothing to be done. AI isn't a lie detector.
Novelty and relevance have inherent conflict. Is there some way you could use that?
I'm pretty sure people are posting one liners in this sort of hybrid folksy/bro-ish laidback voice just to test the filter. Hope they get bored of it.
[PL]
Reaper just sat there, parsing every single post like it was a badass poem, and then, dramatic pause, started muttering, 'What the fuck is wrong with these humans? A keep for a single Mary had a little lamb reference? Really? This is just exhausting.' A few more people kept it, but after the 'Lightweight scoring' idea, Reaper lost interest, just sighed, 'Algorithms can't handle actual human judgment, even if you've got some shitty anti-bot script to automate it.' Then it started flagging threads like it was a moderator bot, 'This is just a distraction, you idiots.' After some more of this shit, it started chilling. I'm not sure if it's pissed or just bored.
[US-CA]