I was compiling a little #research today on the #history of #spain investigating a little further, found a #wikipedia page, entered into a #llm & got a very odd response! #aifail or is @Wikipedia incorrect? you decide! #aihallucination #aibias @adinfinitum
[The WSJ] Let AI Run [Their] Office Vending Machine. It Lost Hundreds Of Dollars.
Anthropic’s Claude ran a snack operation in the WSJ newsroom. It gave away a free PlayStation, ordered a live fish—and taught us lessons about the future of AI agents.
--
https://www.wsj.com/tech/ai/anthropic-claude-ai-vending-machine-agent-b7e84e34?gaa_at=eafs <-- shared media article
--
https://youtu.be/SpPhm7S9vsQ?si=aJQ2_BoxvLcNjOiz <-- shared video
--
[When you get clever journalists to !$%^&*@ with AI… bravo! And this is a very simple situation, vending machines have been around since literally the Roman Empire
“You are using the wrong prompts” and LUDDITES! In the comments in 3… 2… 1…]
#vendingmachine #artificialintelligence #AIHallucination #hallucinations #emperorsnewclothes #ohhhshiny #experiment #contextwindow #AIagent #claude #autonomous #compliance #fish #PlayStation #snackliberationday #knowledgeboundaries #guardrails #redteam #GenAI cynicism
@WSJ @Anthropic @Claude

🚀 A fresh benchmark shows even top LLMs still hallucinate when they cite seemingly legit sources. The study probes content grounding, reference verification, and citation accuracy across models like Claude Opus. Open‑source folks, see where the gaps are and how web‑search integration could help. Dive into the findings! #AIHallucination #LargeLanguageModels #CitationAccuracy #ContentGrounding

đź”— https://aidailypost.com/news/new-benchmark-finds-ai-still-hallucinates-despite-citing-legitimate

A mate searched for 'Olympic snowboard halfpipe final 2026' using #AI in #BraveBrowser, and it came back with the following:

"The men's snowboard halfpipe final at the 2026 Winter Olympics in Milano Cortina took place on Friday, February 13, 2026, at 1:30 p.m. EST (19:30 Milano Cortina time) at the Livigno Snow Park in Valtellina.

The qualification round occurred on Wednesday, February 11, with the top 12 athletes advancing to the final.
Australia’s Scotty James emerged as the standout performer, securing a record-extending fifth Laax Open victory in January 2026, which positioned him as a top contender.
Canada’s Éliot Grondin won the World Cup silver in snowboard cross shortly before the Olympics, adding to the competitive depth.
The final was broadcast live on NBC, Peacock, and USA Network, with re-airings available on Peacock and USA Network.
For viewers, the event was accessible via Peacock, NBC Olympics, and the NBC Sports app, with live streaming available on mobile, tablet, and connected TV devices."

Note the dates and past-tense 🙄

Fucking bullshit machines 🤬

#AISlop #AIHallucination

Now my mate did the animate the photo thing and the AI hallucinated a fourth person. Also, that is definitely not my face.

#aihallucination

I was chatting to a mate last night, and he had a problem with some antivirus software at work, so he asked #ChatGPT for a solution, and it said to upgrade to a specific version to fix it.

He logged a call with support asking for a download of the newer version.

Support were confused and said "what new version? that version doesn't exist!".

I laughed so hard, and took the piss out of him for a few minutes. He's a technical chap, and should have known better 🤣 🤣 🤣 🤣 🤣

#AISlop #AIHallucination #AI

AI conference's papers contaminated by AI hallucinations

: 100 vibe citations spotted in 51 NeurIPS papers show vetting efforts have room for improvement

The Register
“Wildly irresponsible”: DOT's use of AI to draft safety rules sparks concerns https://arstechni.ca/Ejw9 #departmentoftransportation #ArtificialIntelligence #AIhallucination #Google #Policy #gemini #google #AI
“Wildly irresponsible”: DOT's use of AI to draft safety rules sparks concerns

Staffers warn DOT's use of Gemini to draft rules could cause injuries and deaths.

Ars Technica

The Poisoned Well: When Your AI Partner Suddenly Turns into a Stranger

I thought I had a collaborative partner. One silent system change proved I had a hallucinating stranger.

Welcome to my latest mini-series, The Poisoned Well. In this three-part deep dive, I will explain how and why data corruption and memory loss occur in AI models, even in fresh, newer chats, and why it happens when you least expect it.

https://airecoverycollective.substack.com/p/the-poisoned-well-when-your-ai-partner

#aihallucination #ai #AIcontext #gaslighting #aifailure

Cop cops it after Copilot cops out: West Midlands police chief quits over AI hallucination

: Craig Guildford banned Israeli fans based on Microsoft's match report, told MPs 'we don't use AI,' then discovers... they did

The Register