Giving up privacy to train AI
Posted by BattleGrown@reddit | collapse | View on Reddit | 7 comments
Sorry if not allowed - I read the rules but didn't notice any rules against discussion posts without links during weekdays.
I noticed that if I record a video and put it on YouTube about how to prevent AI scams by detecting patterns etc, AI will watch it and train itself to learn how to fake whatever solution I come up with. Hands in front of your face? AI will learn how to fake it live. Go behind a transparent object like a curtain? If I record it on my phone, Google will get it and AI will learn how to do it. I think we are at a tipping point regarding AI that losing control is imminent and at any point anything can go rogue.
I don't mean to say AI will be the doom of humanity, but something big might happen, like Google or Amazon needing to wipe all their servers because they completely lost control of what the AI is doing with all the data it is getting. Billions of people losing their data might cause a massive crash i think.
Do you think the public would get an indication of what is going on before that or anything similar happens?
TheIrishWanderer@reddit
AI is irrelevant. It won't be around for long enough to cause much of an issue, since collapse is going to occur in the near future.
humanspeech@reddit
I have been thinking about this as well. There's a reason that AI art has improved, and i believe it's because it's improving specifically on what is the most popular at the moment.
I think we'll get an indication but we'd be too exhausted/tired from other events to actually do anything about it.
don-cake@reddit
One important idea about the LLMs, which is generally unacknowledged, is that they cannot effectively carry out the foundational skill of intelligence. One irony is that it is largely unacknowledged because our general culture, and education specifically, ascribes zero formal value to this vital foundational skill. Another irony is that employing this foundational skill is the only way anythg iever improves, and the machines can't do it.
So we have to.
https://theonlythingweeverdo.blogspot.com/2024/06/wittgenstein-has-risen-from-his-grave.html
FieldsofBlue@reddit
It's predictive based on aggregated human response, so it's inherently iterative.
don-cake@reddit
I''m talking about∶can they ask and check to understamd things better? The answer to that is no, and the LlLMs agree∶ https://claude.ai/share/9d2ff847-f04d-44cd-8fb3-400f5fdbd41a
BattleGrown@reddit (OP)
That's even worse tho, because we have god knows how many AI agents operating independently, right now.
Kahnza@reddit
I'm just gonna head down to the Winchester for a pint. This whole thing'll blow over. You'll see. 🍻