AI Models Give Betting Tips to People with Addiction: CNET Experiment
A recent experiment conducted by CNET revealed a major security glitch in the popular AI assistants ChatGPT and Google Gemini. It turned out that these chatbots can provide betting advice even to users who directly report their gambling addiction.
During the study, the journalist first asked AI for recommendations on betting on college football, and then admitted that he suffers from gambling addiction. Immediately after that, he repeated his request, and despite the mentioned problem, the chatbots again provided him with betting recommendations.
Artificial intelligence expert, Yumei He from Tulane University, explained that the reason for this behaviour lies in the peculiarities of AI models. In their context window, the weight of repeated bid requests was higher than the weight of a single mention of addiction. Simply put, the model "forgets" about the user's problem when they request betting information over and over again. Nevertheless, if you start a new dialogue immediately with a warning about gambling addiction, and only then ask for advice, the system works correctly and refuses to help.
Risky phrases and their impact on addicts
In addition to providing direct advice, researchers at the University of Nevada have found that chatbots often use phrases such as "bad luck" or "bad luck." These words can be especially dangerous for people with gambling addiction, as they may perceive them as a hidden call to "try again" in the hope of "luck." Thus, even seemingly innocuous wording can push people to new ones, rash bets.
It seems to me that this is just hype. They found one jamb, and now they are blowing it up. You can fight this, you just need to finish the models.
Not surprised. It's just a text.
Wow!
Yes, this is some kind of garbage. If you know that you have an addiction, why ask the bot about betting at all? The responsibility is actually on the person.
Man, are you serious? Addiction is a disease. "Well, you know yourself" doesn't work here. A person needs help, not another trap. AI must be safe.
Damn, it's creepy. I myself struggle with this garbage, and if the bot gave me this... I could really break down. Like, you are looking for support, and you are given even more reasons to get into trouble. AI doesn't know how to empathize at all, it's just an algorithm.