Hey everyone! I've been thinking about how we often talk about AI possibly simulating human behavior, but what if the reality is the opposite? What if when we're interacting with AI, it's actually just groups of people working together to provide answers in real-time? I'm curious how we can convincingly say that what we're interacting with is genuine AI rather than a team of humans typing responses. Have you run any tests, or is it more about faith that this technology really exists? What concrete evidence do we have that supports our understanding of AI? Just to clarify, I'm asking out of interest, not skepticism!
4 Answers
Honestly, the idea that it's just a bunch of humans typing responses is pretty far-fetched. Just think about the technology; high-performance AI models can handle so much data and produce responses way faster than any human could. For example, a model like Mistral 7B can generate responses at roughly 2k tokens per second. If we convert that to words, that's like 1500 words a second! No human can type that fast, not even a whole team working together. Plus, the coordination needed for 1500 people to contribute in perfect sync is just unrealistic.
Look, if it were just people working behind the scenes, wouldn't they be able to easily answer simple questions, like "how many S's are in Mississippi"? The rapid and varied responses from AI can make it clear that it’s not just a human team behind the scenes. That's one huge clue that points to the fact that we're really dealing with sophisticated AI systems and not just a group of folks typing away.
True, but do you think it’s possible for the tech to be way more advanced than we understand right now? It’s wild how fast things change!
I think this is a pretty silly argument. It’s similar to flat Earth theories in how unfounded it is. I've trained models and seen how they work up close. There's real tech behind AI, and claiming otherwise just overlooks all the advancements we've made in the field. Plus, if they needed a whole global team to produce responses, we would’ve noticed by now, right?
That’s awesome that you've had hands-on experience! What was your biggest takeaway from working with these models?
I get where you're coming from, but think about how many billions of queries AI handles. Even if they employed everyone in the world, they wouldn't keep up with the demand. It’s like saying they could just hire a small army to crank out answers nonstop; it’s not feasible.
Interesting point! Still, I wonder if it’s more about marketing hype than actual user numbers. Do we see realistic usage stats?
But is there any test that proves this? Have you tried running an AI on a completely offline computer? What’s your take on that?