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 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.
Instances that don’t vet users sufficiently get defederated for spam. Users then leave for instances that don’t get blocked. If instances are too heavy handed in their moderation then users leave those instances for more open ones and the market of the fediverse will balance itself out to what the users want.
The sad truth is that when Reddit blocked 3rd party apps, and the mods revolted, Reddit was able to drive away the most nerdy users and the disloyal moderators. And this made Reddit a more mainstream place that even my sister and her friends know about now.
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 think I communicated part of this badly. My intent was to address “what is this speech?” classification, to make moderation scale better. I might have misunderstood you but I think you’re talking about a “who is speaking?” problem. That would be solved by something different.
Are they just putting everything on layer 1, and committing to low fees? If so, then it won’t remain decentralized once the blocks are so big that only businesses can download them.
It has adjustable block size and computational cost limits through miner voting, NiPoPoWs enable efficient light clients. Storage Rent cleans up old boxes every four years. Pruned (full) node using a UTXO Set Snapshot is already possible.
Plus you don’t need to bloat the L1, can be done off-chain and authenticated on-chain using highly efficient authenticated data structures.
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.
I’m talking text only, and there are some fundamental limitations in the way current and near future LLMs handle certain questions. They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models. It’s honestly way better than image captchas.
They can also be tripped up if you simulate a repetition loop. They will either give a incorrect answer to try and continue the loop, or if their sampling is overturned, give incorrect answers avoiding instances where the loop is the correct answer.
They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models.
And that is solved just by keeping a non-processed version of the query (or one passed through a different grammar to preserve character counts and typos). It is not a priority because there are no meaningful queries where that matters other than a “gotcha” but you can be sure that will be bolted on if it becomes a problem.
Again, anything this trivial is just a case of a poor training set or an easily bolted on “fix” for something that didn’t have any commercial value outside of getting past simple filters.
Sort of like how we saw captchas go from “type the third letter in the word ‘poop’” to nigh unreadable color blindness tests to just processing computer vision for “self driving” cars.
They can also be tripped up if you simulate a repetition loop.
If you make someone answer multiple questions just to shitpost they are going to go elsewhere. People are terrified of lemmy because there are different instances for crying out loud.
You are also giving people WAY more credit than they deserve.
Well, that’s kind of intuitively true in perpetuity
An effective gate for AI becomes a focus of optimisation
Any effective gate with a motivation to pass will become ineffective after a time, on some level it’s ultimately the classic “gotta be right every time Vs gotta be right once” dichotomy—certainty doesn’t exist.
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.
we have to use trust from real life. it’s the only thing that centralized entities can’t fake
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.
Instances that don’t vet users sufficiently get defederated for spam. Users then leave for instances that don’t get blocked. If instances are too heavy handed in their moderation then users leave those instances for more open ones and the market of the fediverse will balance itself out to what the users want.
I wish this was the case but the average user is uninformed and can’t be bothered leaving.
Otherwise the bigger service would be lemmy, not reddit.
Just like classical macroeconomics, you make the deadly (false) assumption that users are rational and will make the choice that’s best for them.
The sad truth is that when Reddit blocked 3rd party apps, and the mods revolted, Reddit was able to drive away the most nerdy users and the disloyal moderators. And this made Reddit a more mainstream place that even my sister and her friends know about now.
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.
A bot can do that and do it at scale.
I think we are going to need to reconceptualize the Internet and why we are on here at all.
It already is practically impossible to stop bots and I’m a very short time it’ll be completely impossible.
I think I communicated part of this badly. My intent was to address “what is this speech?” classification, to make moderation scale better. I might have misunderstood you but I think you’re talking about a “who is speaking?” problem. That would be solved by something different.
What? I post a lot, but the majority?
…oh, you said LLM. I thought you said LMM.
Also is data scraping as much of an issue?
Reputation systems. There is tech that solves this but Lemmy won’t like it (blockchain)
Do you have a proof of concept that works?
https://docs.ergoplatform.com/eco/reputation-system/
Are they just putting everything on layer 1, and committing to low fees? If so, then it won’t remain decentralized once the blocks are so big that only businesses can download them.
It has adjustable block size and computational cost limits through miner voting, NiPoPoWs enable efficient light clients. Storage Rent cleans up old boxes every four years. Pruned (full) node using a UTXO Set Snapshot is already possible.
Plus you don’t need to bloat the L1, can be done off-chain and authenticated on-chain using highly efficient authenticated data structures.
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.
I’m talking text only, and there are some fundamental limitations in the way current and near future LLMs handle certain questions. They don’t “see” characters in inputs, they see words which get tokenized to their own internal vocabulary, hence any questions along the lines of “How many Ms are in Lemmy” is challenging even for advanced, fine tuned models. It’s honestly way better than image captchas.
They can also be tripped up if you simulate a repetition loop. They will either give a incorrect answer to try and continue the loop, or if their sampling is overturned, give incorrect answers avoiding instances where the loop is the correct answer.
And that is solved just by keeping a non-processed version of the query (or one passed through a different grammar to preserve character counts and typos). It is not a priority because there are no meaningful queries where that matters other than a “gotcha” but you can be sure that will be bolted on if it becomes a problem.
Again, anything this trivial is just a case of a poor training set or an easily bolted on “fix” for something that didn’t have any commercial value outside of getting past simple filters.
Sort of like how we saw captchas go from “type the third letter in the word ‘poop’” to nigh unreadable color blindness tests to just processing computer vision for “self driving” cars.
If you make someone answer multiple questions just to shitpost they are going to go elsewhere. People are terrified of lemmy because there are different instances for crying out loud.
You are also giving people WAY more credit than they deserve.
Well, that’s kind of intuitively true in perpetuity
An effective gate for AI becomes a focus of optimisation
Any effective gate with a motivation to pass will become ineffective after a time, on some level it’s ultimately the classic “gotta be right every time Vs gotta be right once” dichotomy—certainty doesn’t exist.