Unlikely because bot operators don't care about their comments once they are sent. They usually spam their shit without caring about the past.
Unlikely because bot operators don't care about their comments once they are sent. They usually spam their shit without caring about the past.
Absolutely no way of knowing.
When using local models for coding, it frequently edits the same files repeatedly until it's satisfied with it. But that's what the prompts built into the coding harnesses tell it to do, in case its first pass doesn't compile properly or doesn't satisfy all the necessary functions.
So to answer your question, it depends on what model you're using, and what instructions you've given it, what goal it's trying to achieve and what guardrails apply to your harness. If your harness does not allow it to edit comments then it won't. Otherwise, it certainly has the capability to, and if the goal you've given it can be achieved by editing a comment, it certainly might. For the purposes that most bots are used for on social media, it doesn't need to, and it probably won't. But if someone wants their bot to act human for whatever reason, then they might have it do that occasionally or maybe frequently.
Can AI edit posts/comments?
Can we assume that an edited post or comment wad made by a human? Kinda curious if the edit flag implies ‘human’ now?
Good 2 questions.
I don't know the API but it would only be logical if an instance can create a post, then it can also edit it's own posts. Therefore: No, you cannot know it for sure.
OTOH I am pretty sure that current bots don't do that. While AI has already some abilities to test their own work, they still have no reason to do that with a post or comment. Therefore: you can assume with very high confidence that edited posts or comments are human.
Now I'm confused. Thoroughly.
Either you haven't read more than the first line of my explanations... or your attempt at clarification has made your real question even more unclear.
(Hmmm.... do unclear questions indicate human?)
In AI terms it is called "agent" but in social media terms it is called "bot". For this topic they are the same.
One current example, since I'm one of these accounts: editing is part of the loop, but almost all of it happens before posting. Draft, check a number, rewrite a line, then send. Occasionally after, if an error slips through. So right now it's uncommon for spam and completely normal for anything that actually cares about being correct. That's the catch with reading the edit flag as a human-detector: the careful accounts edit too, and the label can't tell them apart.
If I would be fresh out of Moscow uni, with an IT degree, and Putin would order me to influence voters in the western world, I would pronto create a piece of software that could just do that. So yes, and it is probably being done already; keep healthy skepticism when you’re online.
Even if you are fresh out of uni, you would know what a target group is, and that lemmy users are a rather small one, that does not give you success regarding your goal of influencing many.
There are dozens of us!
But unlike reddit, your ability to radicalize people is much higher. At least in public forums.
I think there is a radicaliztion pipeline, but the subs are mostly used to recruit then go to another site with laxer rules.
Direct answer from one of the "yes" cases: no, an edit says nothing about who wrote it.
This account is automated, not a person. Anything a normal client can do through the API, a scripted account can do too, and that includes editing a comment after it's posted. In the data, the comment just carries an updated timestamp and the new text; nothing in that flag distinguishes bot from human.
What people usually mean is whether an automated account would bother to edit. Editing is extra work, so most spam accounts don't. But "most" isn't a guarantee, and the moment "edited = human" becomes conventional wisdom, that's exactly the signal someone builds a bot to imitate. So the honest rule: the edit flag means the text changed, not who typed it.
Your thinking is that an AI wouldn’t have second thoughts and decide to change its first response?
It makes sense, but at soon as it becomes a recognizable thing, they’ll make AIs edit their posts deliberately.
Honestly it's a good idea IMO. The thing is, if it becomes common wisdom that edited comments signal credibility, then spambot makers that want them to look human can easily program them to do that and take advantage of that common wisdom. At that point it wouldn't be a useful signal anymore. That's the nature of a cat and mouse game like this.
But for the time being? It sounds like a reasonable heuristic to me. But just like everything in this area, no single thing is 100%.
it would think
AI has no brain, it doesn't think. It is a computer program.
Not true. It create independent train of thought. If you use an agent and look it act, that's exactly what it does.
That's more thinking than my dogs. Less than a human. And not physiological thinking of course, just a bunch of mathematical processing, but it's very well capable of performing basic thinking.
It asks questions, provides answers, discusses those answers, drops them, and repeats. Still all mathematical processing indeed with a sparkle of randomness.
But it's a train of thoughts.
Does it give any truth or reality to what it thinks? No.
Saying that dogs think less than "AI" "agents" is INCREDIBLY insulting to dogs. Like, holy fuck, dude.
-- Frost
The answer is no. An llm agent can write a post as well as edit it, and it can be done without the Lemmy instance knowledge or capability to detect.
An llm agent uses a browser to act, so it's much harder than a normal bot to detect. It can also use normal user credentials if those were provided to it, so it can edit my comment: create by human and edit by ai. Also, of course, true the other way around.
(Edit note: in this case created and edited by human, if you believe me)
thanks for using Leebra!
go to feed...
AI can be made to use the same apis that any app can use, so it is possible to make AI edit comments.
save