this post was submitted on 07 Aug 2026
322 points (94.7% liked)
Technology
87279 readers
3290 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
What we want is software that behaves predictably. Since LLMs don't do that, we don't want them or their "agents."
For fuck's sake.
To be clear, "normal" people are using LLMs like crazy. This I believe is specifically talking about is installed local agents. Boomers and non tech industry suits are using LLMs left and right and won't shut up about them.
They are not using agents. Most people use LLMs as Google on steroids.
Google on LSD (the hallucinate results)
99% nudify apps
Doesn't help the conversation that some LLM chat systems call minor customization of the prompt making an agent.
Under duress I use CoPilot Chat at work. No, the basic sad customization I did along the lines of "don't be sycophantic, always cite sources, use these specific official documentation sites as sources before searching outside of them and clearly call out when you search outside them" that customization isn't making a damn agent, it's adjusting the prompt and saving those changes you disingenuous marketing fucks!
I've been "adjusting the prompt and saving changes" with Claude code for the past 9 months (it's not ready to have a baby yet), it has been steadily improving over those months, a lot of that seems to be in the models and harnesses provided by Anthropic, but not a small part of it is the customization of the prompt to do what I want the first time instead of making me redirect it constantly.
lol it's true. My mom sends me screenshots of AI chats every fucking day.
Just the other day I was pissed off. My dad missed some deadline to request money from the government. I told him to call them, they might be able to do something anyways. He was like "nah, they'll never do that, why would they have a deadline if they do it after the deadline yadayada"
Then a few days later he told me "oh that gpt thing told me to just try, I might be lucky and they pay out even after the deadline, he was very reassuring, so I'll try calling them!"
Man chatgpt would fucking tell you to try your luck if you're 12 years over the deadline, but god forbid your son tells you the exact same thing and you even consider calling them lol
There was some research recently that showed the same thing. People who are "dug in" on a topic, usually a political topic, will shut down when a person tries to persuade them of an alternative, but are much more open to hear the same arguments from AI and actually reconsider their position.
I believe this is a big reason why X-Ai was so important for conservatives to get off of the ground, they were afraid all the crazy would start wearing off if people asked the "woke" models for the truth.
Yeah this is the deal. I'm a Gen x er and I use Gemini for some tasks - I want a quick id of a component or reference some knowledge chunk to verify something. I have absolutely no use for an agent.
i know some tech people that uses it.
In as far as the Google Home speakers are "Agents" we've had one for several years now. It's good for a laugh once every so often when it gives a wildly random response to something we ask it for. Would we trust it to do anything like lock or unlock the house door? I don't think so, probably not ever, definitely not today.
I actually really like the technology that has been collectively lumped together as "AI". I think it's fairly useful now, and suspect we're at the beginning of the curve for this technology, and it will only continue to improve, even as it becomes more efficient. I understand this is a wildly unpopular stance in these parts, judging by the avalanche of downvotes I get for having this opinion.
What I don't like is feeding tons of information about myself to a giant tech company. I run a local LLM on my phone and another on my (fairly decent) computer, and sometimes I use duckduckgo's front end, which anonymizes prompts.
As for agents, it's pretty much as you say. I just am not ready to have an LLM take action without my direct supervision, because I'm confident that between flaws in the LLM and flaws in how I give it direction, something will go sideways.
Same, it's a tool everyone will use one day, it's just bad today.
Remember the first mobile phones? Sucked dirt, and we're stupidly expensive. Then they got cheaper and better, today they almost give them away and the charge lasts a week and you can phone almost anywhere on the planet.
"AI" will be the same, in some years I guess.
If it does get better, it will be with technology other than LLMs, because LLMs don't get cheaper per unit of usage as usage scales.
I suspect that we won't have actually useful AI of some sort until LLMs get out of the way. They're sucking all the oxygen out of the room right now.
And it's also not like the hardware is improving much either, we've effectively hit the wall performance wise with processors and RAM.
That's literally what all these datacenters are for, we can't scale the individual computer performance up, so instead we build more and more of them.
I bet you can make a cheap "AI" chip for using the trained model, it seems the training is the ruinous thing today.
Remember when they predicted heavier than air flight to be centuries in the future just the week before the wright brothers flew? Me neither I wasn't born then, but it's an interesting example IMO.
Aka reliable technology
It is actually legitimately unclear what to expect of them these days. They can do some dazzling things, and fail miserably at others. The spectrum from “easy” to “hard” is not the same for our brains. And our brains mostly know when something is hard or we just can’t do it. Or at least we’ve had millennia to adapt to the way our brains are. They’re pretty idiosyncratic as well, and both marvelously clever and painfully stupid.
Most people didn't even trust algorithmic automation before agents came along. lol