Why Normal People Aren't Using AI Agents | Hacker News Reader

Why Normal People Aren't Using AI Agents

wired.com

24 pointselo200010 comments

Personally, I don't use them because I feel they're: unsustainable (burn through too many tokens), over-complicated, have privacy/security problems and creating an unhealthy reliance on them to do basic stuff.

Why would most people use them? What do they offer other than dubious facts that they have to check 95% of the time to not get screwed by something important that it got wrong and in the past would have taken a 5 second google search before the top 50 results became a copy-pasted AI answer?

AI may be good for coding, but it is still crap for many generalized topics beyond a 5th grader's questions. The answers are way too simplified and lack the details needed to rely upon it and contains tons of useless filler. Nobody is going to spend a bunch of money just to ask an AI questions that might be given wrong answers and spews out 75% filler that wastes their time.

Go start asking AI about the Ph level needed for different plants. You can get multiple different ranges depending on how you ask the question for the same plant, with repeated generic plant growing info that is not helpful for anyone that already gardens or has grown a plant.

Part of the problem is there’s so much garbage on the internet that an LLM can consume in search of the answer. As humans, we look into source quality: we can tell it by reputation of the website, how carefully crafted that site is, what qualifications the author has, and much more. There are many subtle tells on the internet that indicate source quality to humans, that AI cannot pick up on.

I experienced this first hand this week while searching for a specification on how much oil my car holds. AI gave me different answers multiple times. I manually searched, found an official manufacturer page, and got the proper information. Had I listened to the AI the first time, I would’ve severely underfilled my car.

Sometimes I wonder if I am in a bubble. This has not been my experience at all. A few examples:

We have a sailboat and it has a turbo diesel engine, which is unusual (for a sailboat). Purely mechanical and I understand small diesel engines pretty well. But when we had loss of power and a change in exhaust and it could have been about 4 different things. Based on an array symptoms ChatGPT worked through a few tests for me to do and narrowed it down. It had a consistent and good methodology for working through this.

I am planning a large solar array for our boat. It is challenging because it is a 48v system and that has certain requirements. We have to use an array of smaller panels wired together in groups to get the right voltages. I also have partial shading issues to work out. But there are a lot of other variables too that have to be accounted for. It worked through all this well and came up with a great plan, pointing out things that were not hard to verify but things I did not initially think of. I am pretty knowledgeable about electrical stuff as well and it did not get anything wrong.

Another example is I want to come up with a way to identify a user for a system based on 5 things they could remember. Not facts, things. But how do you make a system that someone can remember (without writing down) 5 things, 5 years from now. ChatGPT came up with a good idea using memory chains that would make it easier for a person to remember.

All of these were insightful conversations that I had with ChatGPT Sol medium effort.

I also find that for answering questions that require some thinking (like how does xyz work) my default is becoming ChatGPT vs Google more and more.

I think that now we are starting to see the hype go bye-bye and the real use cases are starting to emerge. However, the systems that come out must be constructed with care aligned with our world has been constructed (reality) and not how the AI Labs believe it should be (theory).

https://archive.ph/Bj2wj

While I wait for someone to post a non-paywalled version, what's the real overlap of things the average joe wants automated, things an agent can reliably do, and isn't seen as a bit too creepy? I think it's pretty small, and certainly a lot smaller than Silicon Valley thinks. That's still before things like cost, general AI aversion, and the technical confidence to get an agent set up.

I think it could be argued there's a use case for agents to function as a personal secretary, planning, emailing, news briefs, task reminders, those kinds of things.

Probably hasn't taken off because what this solves isn't much of a problem for people without executive dysfunction. Most people simply don't need a secretary, and the tasks you might delegate are often pretty important things that must be done correctly (e.g. you don't want to invite your ex to your wedding, or your boss to your family dinner).

Interconnectivity managers and smarthome managers are about it. People don't want something that can alter things unbidden and unrestricted. But them being able to automatically set up, for example, your phone's Bluetooth connection to a rental car would be appealing to a lot of people. Or being able to adjust the thermostat or start the coffee maker or whatever according to preference of whoever's in the house.

You can just disable js on that page and get text of the article. Works in my Firefox.

Reply on news.ycombinator.com