Contractors say humans review Microsoft Copilot prompts and images
New from 404 Media: when you upload images to Copilot, they're not private. Human contractors are reviewing Copilot users' prompts/images, according to internal docs we got And they're horrified at what they're seeing. Bombarded with requests to make sexual AI images www.404media.co/humans-readi...
Joseph Cox180Crucially these contractors are not hired to filter out offensive content. They are hired to report whether Copilot successfully made a good AI image. So if the request is, make this person's breasts larger, they have to review based on whether it did or not. Insane www.404media.co/humans-readi...
class war puppet - block judge39People horrified to learn the non-consent machine is used most by people who cannot get consent. That about right?
Lyd Ⓥ11It’s infuriating that they hired people to validate the outputs of AI, not filter out obvious inappropriate images. They have the capability to stop these images and instead they’re verifying that these images satisfy the wants of an inappropriate person.
Litbowl6Just finished reading this in the newsletter. Jesus christ. Where is the line on expectation of privacy and "person committing sex crime and should be reported for it" here? (I understand on that level it's not wildly different than a situation with the same images in Google Drive or similar).
_aa_6I have a Google speaker next to my bed and I recently logged on to Gemini and every single question I ask my speaker is logged as a session. It was "OK Google what time is it" over and over thousands of times. I'm glad there's a human keeping an eye on that.
Strange Currencies3I wonder if this is true of the enterprise versions of copilot as well. I definitely know there are people putting sensitive workplace information into copilot on the basis that the enterprise version doesn't use any gathered data to train public models, but if people are reading them anyway, oof.
Olivia Guest · Ολίβια Γκεστ175are we finally going to accept it's all a scam, a "mechanical turk", it's all humans?
Coach Finstock116Not surprised one bit. While it's not the only use, I am willing to bet nearly all people that use AI in any capacity have done prompts like this. Everyone knows it's the CSAM machine. Feature not a bug for the creeps of the world.
Victoria Strauss37More evidence of the hidden ways in which LLMs are powered by (likely exploited and poorly paid) humans
Charlotte Garden35There's no way to make "being bombarded with horrifying images" an okay job -- but this work will inevitably be bad in other ways too. Low pay, unmanageable pace, no agency, etc.
hammancheez26Remember, when you work for Microsoft, we're one big family, and in this family we all work together to make the CSAM generation machine better with every iteration
Paul Bernal17A general reminder: AI is a fundamentally privacy-invasive technology. That's in relation to pretty much every aspect, from the data you upload to the data it scrapes and rips, to the insights it gains and the outputs it produces.
Robert Stribley16Seems like maybe another example of keeping an overtaxed “human in the loop” as a “moral crumple zone,” so your work on LLM itself and what it’s doing can continue unchecked.
[Christopher M. Federico]12"She-Hulk images come up a lot, the person who works on the prompts told 404 Media. Other contractors also wrote they were presented with pro-anorexia content."
Ether Diver9Garfield Big Naturals. Photorealistic. 4K.
Rima I Anabtawi8“Microsoft is explicitly interested in the contractors’ human intuition on which image looks better, according to the documents. “Trust your intuition — when you glance at the two edited images side by side, which one immediately feels like the better edit?…”
Human contractors are reviewing Copilot users’ prompts and uploaded images, according to internal documents obtained by 404 Media. The contractors are also bombarded with users’ requests for sexual AI images.
Kimby🐥1Deepfakes have increased a lot so that's the answer
lanceseamans.bsky.social1Whatever moderation layer exists on the input side for these tools, humans reading raw prompts full of personal images and requests is a detail most people assume is automated away already.
amateurebutter.bsky.social1not artificial and not intelligent; reliant on armies of poorly paid humans uploading and apparently editing human data unpleasant and unnecessary also unsustainable technocracy driven data centre development and liveable planet incompatible bsky.app/profile/404m...
🗽LOLGOP🗽38George Orwell: 'If you want a picture of the future...
izzy kestrel22idk man i feel like we need some kind of rule like “if you haven’t heard a single positive story about a technology after a year, we take it behind the barn and shoot it” how many years have we been putting up with this shit? and it’s still “oops all misery”
JayIsPainting 🍁17Nothing you post to AI is private. Nothing you post to AI is secure. Also grow up.
Snigdha17Good to know we are building energy and water guzzling data centres, with all the noise and environmental damage caused so Copilot users can create materials to assist bashing the bishop. So worthwhile... 🙄🙄🙄
Developer Soup3I'd have thought the Facebook video moderators experiences would have already taught us that looking directly into the void isn't healthy.
Tyler Bradley3"Schedule a meeting with Lynda on Thursday at 2pm . Include the action items spreadsheet from last week's standup and a word document summarizing our related emails from this week. Also use her teams profile pic and SHOW ME HER BEWBS.". . The future
ScienceFemme2Not only is this incredibly disturbing, but it gives the lie behind AI: it's not independently learning, it's being corrected, updated, and trained by armies of remote human workers in real time. Anyone who falls for the "AI will replace us" myth is a doofus.
