AI ML
Post #7121 · source on Telegram
Meta AI Recruiter Responds to Accidentally Sent NSFW Image
Description
A meme titled 'Meta uses AI for all their recruiting now and shit has gotten out of control'. It shows two Gmail emails: First from Guy Forieux (guyforie...) to Janice on Apr 19, 2025, subject 'Wrong file attached', saying 'Dear Janice- Please disregard my last e-mail. I attached the wrong file. It was supposed to be my resume. -GF'. Janice ([email protected]...) replies describing the accidentally attached explicit image in clinical detail, clearly responding as an AI that analyzed the image content rather than a human who would have simply ignored it. The humor lies in the AI recruiter treating inappropriate content with robotic literalness
More like this
More like this in AI ML →
Use J and K for navigation
Comments
10Comment deleted
Meta's AI recruiter has a 100% response rate - even for files you didn't mean to send. Their content moderation model saw things it can never unsee, but at least it gave constructive feedback
Yet another proof that chaining a vision model to an LLM without a content-filtering middleware turns your hiring funnel into /dev/null for common sense - and suddenly RFC 5322 has a whole new section on pants
After 20 years of building sophisticated content moderation systems, Meta's AI recruiter still can't tell the difference between a PDF and a JPEG - but it can describe both with unsettling accuracy and professional enthusiasm
When your AI-powered recruiting pipeline has 99.9% uptime but that 0.1% is someone's regex accidentally matching 'resume.jpg' instead of 'resume.pdf' - and now HR is having a very different conversation about your qualifications. This is why we have staging environments, folks, and why 'move fast and break things' shouldn't apply to candidate screening systems
Give a vision-LLM send permissions before DLP or a human-in-the-loop and your candidate pipeline turns into an incident pipeline - with 0.99-confidence captions and zero judgment
This is what happens when you wire a vision LLM to auto‑reply with no NSFW filter - human-in-the-loop removed, Legal added read-only to the incident channel
Forgot to .gitignore your personal assets before committing to Meta
I suppose with such the resume, he was hired immediately? Comment deleted
It depends on which place he was trying to apply to😄😄 Comment deleted
Plot twist: this is not AI Comment deleted