That google AI demo with the leather jacket was a total mess and nobody's talking about the real problem

I saw that whole Google Gemini demo last week where the guy asked about leather jackets and got that weird response about not being able to help with that. Everyone's laughing about it being woke or whatever but I think people are missing the real issue. The AI clearly didn't understand context or sarcasm at all, it just had some scripted filter that blocked certain topics. I run my own electrical shop so I deal with software that makes dumb decisions all the time. Last year our inventory system flagged conduit as a dangerous weapon and locked up for 3 days. My point is these AI safety guards are way too blunt, they block everything instead of actually understanding what's being asked. Has anyone else noticed these systems getting more restrictive and less useful at the same time?

2 comments

2 Comments

the_charles7d ago

Read an article about this in Wired that put it pretty well - they called it "safety theater" where companies look like they're fixing problems but just make things worse. The issue is these filters treat every request like it's coming from someone trying to break the rules, not a regular user asking a normal question. Remember when YouTube started banning comments about "hurtful speech" and ended up blocking tons of harmless videos? Same kind of thing happening here. Did the demo actually show the filter blocking something about leather jackets specifically?

taylor_fox7d ago

Did you catch that Wired piece too? I think they nailed it with the "safety theater" thing. It's like these companies are so scared of getting yelled at that they just slap on these dumb filters without thinking. I heard a podcast where a former Google engineer talked about how their internal testing showed these guards blocking like 40% of normal questions by mistake. The leather jacket thing isn't about being woke, it's about an AI that literally cannot tell the difference between a joke and a threat. My buddy works at a tech startup and he said their chatbot blocked someone asking about "how to cook chicken" because it flagged the word "cook" as potentially violent. That's just lazy programming, they throw these massive filters on instead of teaching the AI to actually understand stuff.