|
That's not what I meant, nor is "touching grass" necessarily relevant. I use these tools at work. If I talked to other people instead of a model every time (or even just some of the time) I talk to a model, I'd be tanking everyone's productivity by constantly disrupting them, and would get worse quality responses pretty much guaranteed, if I wouldn't be just ignored outright. This isn't to say I stopped talking to colleagues either. I reach out to them about the same amount. The time budget I took away from was the one I used to spend manually doing things, reading docs, and so on. So your entire mental model of this is just outright false there. On a personal level, I traded off entertainment time (though for me, interacting with ChatGPT is also just a form of entertainment). Instead of staring at YouTube, I pitch silly thought experiments, explore topics, ask difficult-to-search-for questions. So once again, I did not trade off social time, meaning what you propose simply does not apply whatsoever. I do not share memes or catch up on life with it, which is what friends are for. It has no sense of actual humor, has no life of its own, and I'm not into roleplay like that. Would be just weird. There are people who engage pathologically with LLMs, but that is neither the premiere use nor the goal. I'm sure there are also situations where people will actively approach a model over a person, but I do believe people should be able to self-regulate that decision. It gives people leverage to choose the extent and venues of their social engagement. If a topic has a community you do not wish to interact with, you can finally choose not to, while still reaping any prospective benefits (information). This can be a bad thing, but I think it's easy to see that it is also very much good. For example, there were times where I felt more compelled to interact with ChatGPT than I felt leaving a comment here on HN. And frankly, I'm pretty sure I was way better off, and maybe the people here were too. Or a connected but different example, there were HN comments I took to heart pretty bad. Relitigating them with ChatGPT enabled me to get over them in-depth, and even consider different viewpoints properly, without having to be the resident punching bag of some asshole. It's difficult for me to see these as anything but a net improvement. There is something to be said about interactions in one's life becoming overly structured. I think a reasonable case could be made that in less structured lives, where things just kind of happen, you get more random interactions, and it is through those random interactions that you gain valuable experiences and relationships, even if there's a lot of cruft in addition [0]. That in our effort at optimization, we're making our lives inhuman to live, for the same reason people are crapping on LLMs being statistical (unreliable, imperfect) rather than hardwired (reliable, perfect). But that implies there is an optimal life, which I'm not sure exists: specifically because of this local vs global optimality tension. In complex stateful systems, you're guaranteed to hit scenarios where locally correct and globally correct behaviors are irreconcilable [1], unless you meticulously construct them such that this cannot happen, after all. [0] like https://en.wikipedia.org/wiki/Small-world_network [1] like https://en.wikipedia.org/wiki/Tragedy_of_the_commons |
Read the GPT Live announcement. Why do they want to make it more life-like, if not to encourage the pathological engagement? Why does the model say "thank you for that insightful question"?
It is not "just a tool", that is not the goal. It is designed to be a replacement for people. Even in the most charitable "LLMs are just a tool" reading. The value proposition is to have fewer people (dramatically fewer: the "one person unicorn" with fleets of AI agents!) and for those people to spend more time talking to bots. Replace your junior employees with bots. Replace your customer service people with bots. Replace a bored teenager behind the Wendy's counter with a servile golem whose smooth plastic face can easily be wiped down when spit on.
Their metric is engagement, it is how engaged are you and how long do you spend using the application. They will do absolutely anything to get that number up. Remember, after the GPT-4O debacle, when Altman talked about implementing sex bots? That's a shark, smelling blood. Something like 30% of under 18's have used AI chatbots for "companionship," a big subset of those for ERP.
> Would be just weird.
I was a weird kid who spent 10-12 hours a day on the computer, back when computers were primarily low-res screens and boring text. Now everyone is spending 10-12 hours a day on the computer. These things change in a single generation and I'm sure you know as well as I do that boy does it happen fast.
You sound like me: an older, healthy, well-adjusted person who can understand what's going on when the model says "that's a really insightful question!" Most people are not. They will not understand when it's an AI video clone of Chris Helmsworth.
> there were HN comments I took to heart pretty bad
I bet it made you feel alive.