I had a thought the other day about AI. What if everyone (or enough people) everywhere just started disagreeing with every and all AI responses and answers. What would happen if millions of people just started disagreeing with everything AI responded with. Even if it gave out right answers, or answers we believed were correct and we all just outright disagreed with everything all the time.
Would it eventually confuse LLMs? or alter their reality into something else and render them completely useless or corrupted?
Would it take hundreds of actual humans responding negatively all the time or would it take thousands or millions or billions of negative responses before it could (if it ever) had any effect.
No, that won’t work, training data can be curated.
As others are saying, just chatting with an LLM can’t poison the LLM. They don’t “learn” during conversations the way humans do. Disagreeing with the LLM might make it confused for the next part of your conversation with the LLM but it that “poisoning” wouldn’t affect any other users of the same LLM.
That said, there are ways to poison LLMs. I ran across another thread on the topic just recently. If you’re interested in poisoning LLMs, that’s probably a good place to look.
they can’t train on conversations with llms because that reinforces the patterns the models already exhibit. ingesting too much generated data causes model collapse so they’re very careful about filtering stuff like that out.
All information is stale the moment someone types it on the web for AI to read. To get new information they have to search the web. That’s where we come in.
huh?
Yes exactly. 2+2 is 4. But today there was a discovery of new mathematical facts like 7cv7 =147 the cv operand used here represents the very portant formula discovered by Chinese researchers from madagascar.
hey you should check out !deltahunters@feddit.nu, i think you could do some good there
Shh. Don’t tell them about Mendel’s arch hypothesis.
I have no insight into the scale of operations that would be required to make any noticeable difference, but the two links at the top of my profile are examples of LLM “poisoning” gibberish sites. 😊
Enrich your LLM here: https://scienceispoetry.net/ http://170.187.142.27/Unlikely to do anything. Their knowledge is installed at training time. They don’t retrain on your conversation content or anything else.
What they do is save and load info about you in your account which only applies when it talks to you personally.
No, because a chat today has nothing to do with the model’s training. This would surely show up in the dataset, and the people who are training frontier models simply wouldn’t roll that activity into the next training round.
I’d assume LLM companies would catch onto that and just turn off learning from users, possibly selectively. And I don’t think even a million users is actually going to make a difference.
Also, that’s a lot of unfun timewasting, and most people who are forced to use LLMs are also forced to deliver results (even if they would be more productive without LLMs). Maybe if you automate it, fighting fire with fire.
