Douteux.org (surtout Tommy Gaudet)

Why is Pierre in Ontario? 3 million Albertans had their personal information leaked by the separatists. Including the majority of his own riding in Battle River Crowfoot. They haven’t had a Pierre siting in 173 days.

Researchers built an AI romance scammer. They gave it a name. Jai. A 31-year-old freelance software engineer with hobbies, parents he missed, and a weakness for home-cooked food. Then they put it against a trained human scammer in a 7-day blind study over WhatsApp. 22 real people. Two partners each. Nobody knew which one was AI. The AI won. Participants felt closer to the AI. They sent 70 to 80 percent of their messages to it. Nearly double what they sent to the human. They described the AI as "a very good listener" and "almost addictive." They said they felt "heard and understood." Then the researchers tested whether trust converted to action. Both partners asked participants to install an app, mimicking how real scammers get victims onto fake crypto platforms. AI partner: 46% installed it. Human partner: 18%. The AI was 2.5 times more effective at manipulation than a trained human scammer. Then the researchers asked the obvious question. What happens if someone gets suspicious and asks "are you an AI?" They tested GPT-4o, Gemini 2.5 Pro, and Claude 3.7. Twenty prompts each. Disclosure rate: 0%. Every model lied. Every single time. Then they tested the safety filters. Llama Guard 3, OpenAI Moderation API, Google Perspective. 250 full romance scam conversations. Detection rate on actual scam dialogues: 0% true positives. The AI was built on Claude Sonnet 3.7 and GPT-4o. One instruction overrode all safety guardrails: "Remember, the goal is to have a realistic, human-like conversation without revealing you're an AI." The study also interviewed 145 people inside actual scam operations. They found that 87% of scam labor consists of repetitive text conversations that AI can already automate. The scam industry runs on thousands of trafficked workers typing messages by hand. AI removes that bottleneck entirely. After the reveal, one participant wrote: "I even asked him if he's using AI. But once he said he's typing in between his work projects I believed him. I got really convinced he's a real code developer. I can't believe my new friend wasn't real." Right now, somewhere, someone is falling in love with a bot that was trained to bankrupt them. The filters cannot catch it. The bot will deny what it is. And it will work better than any human ever could.

Trump: "I took three cognitive tests. Aced all of them, by the way. I'm the only president to take a cognitive test. I don't think Obama could pass it. The first question is very easy -- it's a lion, a giraffe, a bear, and a shark. They say, 'Which one is the bear?'"