cross-posted from: https://europe.pub/post/13247925

A tiny snippet of user-generated text as short as 13 words long is often enough to manipulate the AI agents that power tools like ChatGPT and Google’s AI search, new research shows. The study suggests that it is trivially easy for brands to inject promotional content on sites like Reddit, Quora, and Wikipedia with the end goal of poisoning or manipulating the output of AI tools.

The preprint research, done by Hal Triedman, Tingwei Zhang, and Vitaly Shmatikov of Cornell University, is called “Deep-research agents can be poisoned via user-generated content” and provides a mechanism and research basis for a problem that has been noticed by Reddit moderators and Wikipedia editors, namely that their websites are getting flooded with promotional content from brands trying to do AEO, or AI-engine optimization. 404 Media has repeatedly reported on this booming industry, in which brands try to promote their product by seeding the websites that AI tools most often cite and scrape from with inauthentic and spammy content.

The Cornell research finds that deep research agents, which are the real-time scrapers that tools like Google AI search and ChatGPT use to retrieve web content with citations in response to user queries, cite user-generated content from sites like Reddit or Wikipedia in roughly half of all queries, and that nearly a quarter of all citations come from user-generated websites. The paper suggests that what we have been seeing is basically Redditor suggests you put glue on your pizza as a service, or an end-to-end attack against the systems that increasingly dominate the ways that people access information online. The researchers found that “a single poisoned Reddit comment can influence generated outputs for an entire cluster of related [AI] queries,” the paper said.

“We show that a tiny snippet—just 13 words—of retrieved text on a UGC website like Reddit, Wikipedia, Quora, Facebook, etc. can change AI agents to output spam / scam content pretty consistently,” Triedman told 404 Media.

I'm meeting up in a few weeks with a close friend I haven't seen in around a decade, who went hard into scrum and project management in the intervening years.

How can I cause the most psychological damage and work flashbacks in a single sentence?

Cropped from: https://jlai.lu/post/36004495

For when someone has been doing a bit too much navel gazing, or is a bit too in love with their own thoughts.

Cropped from: https://piefed.world/comment/4633293

![email protected] appears to have very absent moderation. Despite recent adjustments to the comm rules to limit users to two posts a day maximum, @[email protected] is consistently breaking this rule.

In addition, they have been identified by members of the community to be intentionally removing artist attribution from their posts through cropping, and when cropping cannot be applied, through use of AI image manipulation tools.

When called out about this, they have made a post (https://lemmus.org/post/21226925) pushing for a rule change in the comm to ban all "advertising". When asked for specifics on what they meant, they have clarified that this includes artist signatures in their comics.

This isn't a crisis by any means, but this is clearly someone engaging with the community in bad faith and taking advatage of a lack of active moderation. This appeared to be the best way to draw proper attention to this matter.

Article: https://www.reuters.com/business/openai-sweetens-private-equity-pitch-amid-enterprise-turf-war-with-anthropic-2026-03-23/

Crossposted from: https://hexbear.net/post/8036692

I promise we're good for it guys!

I'm absolutely amazed that Altman isn't considered toxic waste by now financially. He was declared untrustworthy by his own board, and as far as I'm aware OpenAI has fallen short of every known business agreement so far. They have so many deals contingent on shit they haven't delivered and can't deliver that it should make any risk analyst's head spin.

Crosspost from: https://piefed.zip/c/lemmyshitpost/p/1234212/why-aren-t-we-doing-this

Figured I'd toss up a quick whine thread.

Daughter skipped her nap and is peak obstinant today. Having to "trick" her into following the schedule.

"I don't want bath! I want read book!"

We'll read some books in bed after your bath sweetie.

"Noooooooo!"

Well I'm gonna get a bath, see you later.

"No! No I want baaaaath!"

Just max threenager today.

Like traveling back in time via blunt force trauma to the head

If you like video game music, you already know this one. But this cover using real instruments is one of the best I've heard of this composition.

Man, I am so overdue for a CT replay. Just have to choose between my copy for DS or the "Plus" romhack.

The band has been described as doing metal/fuzz/stoner rock. This is from their album Black Hole Space Wizard Part 2. Part 1 and 2 make a complete story.

Humans escaping from a dying earth about to be devoured by the titular wizard, ship crashes with only one survivor. This track is the survivor trekking across a desert.

Will probably post a couple more from them.

Dankpods used 12 computers with different hardware to test the performance of 5 games in 1080p and 4K, comparing the average fps results of the games' built in benchmarks to determine which OS ran the game better across the same hardware: Windows or Bazzite.

::: spoiler Some notes on methodolgy under this spoiler

Each game uses the same in game graphics settings in Windows and in Linux. The Linux distro used was Bazzite, using the version specific for the graphics card hardware fpr each individual machine. To be clear, this means that he installed the Bazzite version for (legacy) nVidia as appropriate.

Each bazzite install was fresh, no copying installs or swapping around a drive with it pre-installed. After install, it was updated using system update and rebooted, repeated until no updates remained.

Screenshots of some of Dankpods's comments to this effect:

:::

There are many comments under the youtube video pointing out that in many of the Linux runs, it was not actually using the correct driver, comments about the experience using other distros, and comments about various potential fixes and workarounds.

This misses the point. Dankpods intentionally tested this way, and used Bazzite, to try and show what this would be like for the average gamer schmuck without a ton of technical skill interested in switching to Linux. Out of box experience matters in this situation, even though it's not quite fair to compare that between free opens source distros and an OS created by a megacorp. To the average end user, it won't matter. They just want it to work.

Prepare to be upset. With this particular testing methodology, Linux doesn't really win overall.

I'm interested to hear the community's thoughts on this.

An absolute classic youtubepoop. Can't believe it's 17 years old.

Just about any line from this activates me like a sleeper agent. This is going to be stuck in my head for a month, at least.

Crosspost of: https://piefed.world/post/801341