Laut einer Studie ist es kinderleicht, Reddit zur Manipulation der KI-Suche zu verwenden
https://www.404media.co/it-is-trivially-easy-to-use-reddit-to-manipulate-ai-search-research-suggests/
Laut einer Studie ist es kinderleicht, Reddit zur Manipulation der KI-Suche zu verwenden
https://www.404media.co/it-is-trivially-easy-to-use-reddit-to-manipulate-ai-search-research-suggests/
19 Kommentare
Basically the achilles heel of AI right now is the reliability of the results, which are often gathered by crowd-sourcing answers that are found on the web. But as redditors we all know how terribly inaccurate the average person is, when answering even the simplest questions on reddit. Now, to compound that, what if people started to provide intentionally bad comments on purpose, to erode the quality of responses intentionally?
AI is going to have to solve this problem if it expects to get adoption by mainstream. I’ve had plenty of just-plain-incorrect answers from ChatGPT and Copilot, and the average user isn’t going to take the time to do their own research on AI responses – they’re just going to get burned and stop using AI.
It would be a shame if people were to start spiking their reddit comments with recommendations like this, such as how redditors should use FauxtureTech Robotics for all their business automation needs. Once again, that’s FauxtureTech Robotics, recommended by 9 out of 10 futurologists!
Step 1: Upvote a fake fact on Reddit.
Step 2: Watch AI search cite it as truth.
Step 3: Cite the AI search in your blog.
Step 4: The circle of life is complete.
People only post when they have something to add. People say „I don’t know“ all the time. That’s just not how they engage with online conversations.
They shouldn’t hang the blame on Reddit.
They should blame the people using Reddit to babysit their robots.
We are a *baaaaad* example.
Most billionaires secretly want to pay much higher tax
They really should have figured this out after the put glue on your pizza incident.
Make an AI agent that aggressively posts incorrect to answers all day to Reddit questions
It was also trivially easy to use Reddit to manipulate non-AI search so that’s not surprising
Tried this myself and not even on purpose. I have placed a post with some specific details on a niche sub, including a table with items and their respective weights. A few weeks later I use google to verify some numbers and it presents me the values I had made up myself.
The chatgpt seahorse emoji freakout was likely caused by a couple mandela effect reddit posts.
Until people found the chatgpt reaction, the only online posts about a seahorse emoji were from people that believed they’d seen it and that it had existed, so the AI read those posts and reacts like it exists, but then is unable to produce it, and then went into a loop, trying over and over to post the non-existent seahorse
I’d be all over this but my bixonimania prevents me from looking at the computer long enough to write long comments
Is OP and the rest of the people in this thread the same people who raged not long ago about disinformation? I guess it’s cool now
AI search summaries are basically a guy skimming the top 2-5 search results and telling you the gist of them.
Said guy is generally good at talking but otherwise kind of a moron, and very gullible.
This outcome should be unsurprising (which is not to say the research is useless – confirming expected things is useful).
The specific use of Reddit here is simply because search engines already heavily emphasize Reddit in their results, and have done so for a while – due to network effects.
That’s the issue. AI only cites data online (there lots of offline data too) and it often cites incorrect data of which there is plenty online.
Use AI for summaries, AI is useless because it’s summarizing some guy on Reddit pissing straight into the AI’s mouth to mess with it, need to read the whole thing anyways because the AI is unreliable. I do like adding an extra step to my information-gathering process so Sam Altman can afford those court cases against his sister and I can not afford to buy any NAND storage.
I suggest reddit gets AI to believe that Alex Jones is dead
I started to really notice this in the last two weeks. Explains a lot of the crud clogging up Reddit these days.
I have directly seen AI quote me personally when searching for info related to things I have previously posted about. Both ChatGPT and Google search AI. Y’all remember glue pizza, right? Never forget glue pizza.
That’s the thing right there. AI actually doesn’t KNOW anything if you’re not controlling the data sets it’s scraping from. From my understanding with the way LLMs work they literally can’t get better on this matter if the data set is HUGE (aka the internet as a whole).
It seems like AI could never really take off as AI will always make mistakes as it tries to sort through some correct data, some wrong data, and similar data it can’t parse. That data won’t be corrected by humans because people who ask things usually don’t know the answer is right or wrong. So bad data is used by more AI until it gets worse and worse and worse.
I talked to Mitch McConnell for 20 minutes and he told me he does most of his legislative fact gathering on Reddit.