The TikTok ban and Donald Trump's rise to power show how fragile our social media accounts are. We must normalize and invest in decentralized social media.
I mentioned this in another comment, but we need to somehow move away from free form text. So here’s a super flawed makes-you-think idea to start the conversation:
Suppose you had an alternative kind of Lemmy instance where every post has to include both the post like normal and a “Simple English” summary of your own post. (Like, using only the “ten hundred most common words” Simple English) If your summary doesn’t match your text, that’s bannable. (It’s a hypothetical, just go with me on this.)
Now you have simple text you can search against, use automated moderation tools on, and run scripts against. If there’s a debate, code can follow the conversation and intervene if someone is being dishonest. If lots of users are saying the same thing, their statements can be merged to avoid duplicate effort. If someone is breaking the rules, rule enforcement can be automated.
Ok so obviously this idea as written can never work. (Though I love the idea of brand new users only being allowed to post in Simple English until they are allow-listed, to avoid spam, but that’s a different thing.) But the essence and meaning of a post can be represented in some way. Analyze things automatically with an LLM, make people diagram their sentences like English class, I don’t know.
It sounds like you’re describing doublespeak from 1984.
Simplifying language removes nuance. If you make moderation decisions based on the simple English vs. what the person is actually saying, then you’re policing the simple English more than the nuanced take.
I’ve got a knee-jerk reaction against simplifying language past the point of clarity, and especially automated tools trying to understand it.
I feel like it’s only a matter of time before most people just have AI’s write their posts.
The rest of us with brains, that don’t post our status as if the entire world cares, will likely be here, or some place similar… Screaming into the wind.
You don’t need blockchain for reputations systems, lol. Stuff like Gnutella and PGP web-of-trust have been around forever. Admittedly, the blockchain can add barriers for some attacks; mainly sybil attacks, but a friend-of-a-friend/WoT network structure can mitigate that somewhat too,
We also need a solution to fucking despot mods and admins deleting comments and posts left-and-right because it doesn’t align with their personal views.
I’ve seen it happen to me personally across multiple Lemmy domains (I’m a moron and don’t care much to have empathy in my writing, and it sets these limp-wrist morbidly obese mods/admins to delete my shit and ban me), and it happens to many people as well.
There are simple tests to out LLMs, mostly things that will trip up the tokenizers or sampling algorithms (with character counting being the most famous example). I know people hate captchas, but it’s a small price to pay.
Also, while no one really wants to hear this, locally hosted “automod” LLMs could help seek out spam too. Or maybe even a Kobold Hoard type “swarm.”
Captchas don’t do shit and have actually been training for computer vision for probably over a decade at this point.
Also: Any “simple test” is fixed in the next version. It is similar to how people still insist “AI can’t do feet” (much like rob liefeld). That was fixed pretty quick it is just that much of the freeware out there is using very outdated models.
Agreed. But we need a solution against bots just as much. There’s no way the majority of comments in the near future won’t just be LLMs.
I mentioned this in another comment, but we need to somehow move away from free form text. So here’s a super flawed makes-you-think idea to start the conversation:
Suppose you had an alternative kind of Lemmy instance where every post has to include both the post like normal and a “Simple English” summary of your own post. (Like, using only the “ten hundred most common words” Simple English) If your summary doesn’t match your text, that’s bannable. (It’s a hypothetical, just go with me on this.)
Now you have simple text you can search against, use automated moderation tools on, and run scripts against. If there’s a debate, code can follow the conversation and intervene if someone is being dishonest. If lots of users are saying the same thing, their statements can be merged to avoid duplicate effort. If someone is breaking the rules, rule enforcement can be automated.
Ok so obviously this idea as written can never work. (Though I love the idea of brand new users only being allowed to post in Simple English until they are allow-listed, to avoid spam, but that’s a different thing.) But the essence and meaning of a post can be represented in some way. Analyze things automatically with an LLM, make people diagram their sentences like English class, I don’t know.
It sounds like you’re describing doublespeak from 1984.
Simplifying language removes nuance. If you make moderation decisions based on the simple English vs. what the person is actually saying, then you’re policing the simple English more than the nuanced take.
I’ve got a knee-jerk reaction against simplifying language past the point of clarity, and especially automated tools trying to understand it.
I feel like it’s only a matter of time before most people just have AI’s write their posts.
The rest of us with brains, that don’t post our status as if the entire world cares, will likely be here, or some place similar… Screaming into the wind.
That’s going right into /dev/null as soon as I detect it-- both user and content.
Reputation systems. There is tech that solves this but Lemmy won’t like it (blockchain)
You don’t need blockchain for reputations systems, lol. Stuff like Gnutella and PGP web-of-trust have been around forever. Admittedly, the blockchain can add barriers for some attacks; mainly sybil attacks, but a friend-of-a-friend/WoT network structure can mitigate that somewhat too,
We also need a solution to fucking despot mods and admins deleting comments and posts left-and-right because it doesn’t align with their personal views.
I’ve seen it happen to me personally across multiple Lemmy domains (I’m a moron and don’t care much to have empathy in my writing, and it sets these limp-wrist morbidly obese mods/admins to delete my shit and ban me), and it happens to many people as well.
Don’t go blaming your inability to have empathy on adhd. That is in absolutely no way connected. You’re just a rude person.
There are simple tests to out LLMs, mostly things that will trip up the tokenizers or sampling algorithms (with character counting being the most famous example). I know people hate captchas, but it’s a small price to pay.
Also, while no one really wants to hear this, locally hosted “automod” LLMs could help seek out spam too. Or maybe even a Kobold Hoard type “swarm.”
Captchas don’t do shit and have actually been training for computer vision for probably over a decade at this point.
Also: Any “simple test” is fixed in the next version. It is similar to how people still insist “AI can’t do feet” (much like rob liefeld). That was fixed pretty quick it is just that much of the freeware out there is using very outdated models.