Adding my 50 hours~

For the confused:
GenAI doesn't give you the answer, it generates a response that looks like what the answer to your question would look like.

That's why it gives an invalid link- because the answer WOULD contain a download link, in this case!

Close- match it with a cage, still on display so everyone can see. I'll bring a leash.

Good luck everyone!

Now, should we even manage to build a true AI? THAT would work, since that's basically just a digital person doing the checks. But LLMs will never produce a true AI, no matter how complex they get.

As for using AI for age-gating, that sadly wouldn't work. Again, AI don't understand anything- it's a machine with access to a lot of data and algorithms dictating how to use that data to generate responses. They're much easier to trick than older methods of age verification.

LLMs have also been known to do what any actual person would know is the worst option in a given situation, like encouraging people to do drugs, or when someone expresses suicidal intent... telling them to do it. Because AI doesn't UNDERSTAND anything- it's a response replication machine.

AI are also unreliable. The basics are that they have a goal and ANY action to achieve that goal is acceptable. For instance, an AI for testing code doesn't have the goal of testing code- it has the goal of giving positive test results. As such, some such AI have been caught faking tests.

Note: AI have not shown a capacity for empahy, they are LLMs that recognise certain patterns (What appears to be a sad story) and their training directs them to respond in a certain way.
An AI exclusively trained on 4Chan posts would respond with abuse to the sad story.

When you restrict the more legitimate access, the restricted item gets shared through other means. That increases its "coolness" factor and increases the chance your kids will be exposed to it as a result. "They don't want us to see this, let's find out why".
Proper age-gating is the solution.