I said it two years ago already: An LLM is the perfect user interface.

The problem though: It will make it a lot easier to do harder things which will devaluate jobs.

These laymen presumably don't even know what a zip file is, so what are they going to prompt? Let's assume they even know in the first place that they could ask the LLM to "Open this file for me". They also presumably don't know what a filesystem is, or where their file is located within that filesystem. "Open this file that I just downloaded from my browser". They have a billion random zip files strewn across their downloads folder they never, ever clear, or as seems to happen often with the tech non-savvy, they somehow manage to move the file into some random directory somewhere. They're now waiting 10 minutes for the LLM to scan their entire system for MyTaxes(327).rar.zip, only for the "agent" to shit itself upon encountering some sort of permissions block, or for the unzip toolcall to fail, or myriad other reasons why this could fail. Now what? Do you think laymen like these have the patience to guide the stupid bot do to what they want via verbal instructions, when they can't even be bothered to double click a zip file which on any modern system in the last ~10 years will simply open the zip file?

People in tech severely underestimate laymen's computer acumen, and they SEVERELY underestimate people's willingness to put up with software, there is no universe in which they will patiently sit there while the LLM goes through the "thinking" process to operate their computers.

Hell, I work in a tech company and I am often dumbfounded (in a non-judgy way!) at how little people understand their systems or how any of it works despite spending countless hours on their laptops every single day. I don't even mean the nitty gritty technical details, just stuff like "Where did I put that file", "How do I open this mp4 you sent me?", "What do I do with this .zip file?".

Maybe not everyone (yet) but your neighbours clearly do need it.

And if one of my neighbours ever asked me to help them open a zip file, I would not leave before I'm absolutely sure that they can do it themselves in the future.

There's so much ad money to go around. It currently fuels Meta and Google. Let's say Anthropic and OpenAI enter the field and take a slice. That's a fine business, sure. But does it justify the state of the market?

And at what cost to the underlying quality of the product and its utility, and will that cost be acceptable to the end users?

Then all the real AI work I'm using my local setup on dedicated hardware for large local models. I have usede my Copilot pr sub a few times this month to check work but it's either:

1. I ask about bugs and security fixes, glaring holes, and other checks before calling the work complete and posting a release to my persona github. 2% credit usage over 20 minutes and useful small changes. 2. I ask about information in a repo I'm using, like GUFO or Halogen-flash-next, and some key info. 35% of my monthly premium credits gone before the one chat completes.

How can 20 minutes on a whole custom project looking for bugs and issues use almost nothing, then a question about a repo eats 30%+? I don't even have budget to learn how I failed as 2 more tests and I'm burned on premium requests for the month (in minutes).

So I abuse the free options for everything I can. Draft scripts/sql/etc, stage prompts or handoff/etc files, ask questions or research something and pass on to vs code/copilot chat/etc like I run at work (with much more credits through the enterprise plan).

If I were to guess, a lot of people, particularly older people, see computer use outside the web browser as an amateur might a chop-saw; a dangerous tool not to be used by inexperienced hands, so an AI that can also directly do the tasks for them might solve that problem (assuming it came preinstalled or someone else did that for them). Though then comes the question of whether they'd be willing to pay for it

Exactly! I noticed this when I moved out of my parents' house, when suddenly my mom became an electronics whiz who can hook up VCRs and TVs just fine. I always thought I was the smart person in the house, but now I realize I was just taken advantage of.

Compare flights? You have to assume it is objective and not tainted by corporate bias or backroom deals(aka ads and sponsored content), unless you selfhost and find an objective model.

Search Wikipedia? You have to constantly check that it is giving you straight facts instead of being constrained by its community and safety guidelines.

The problem with so many of these usecases is treating AI as a super intelligent Vulcan who is always logical and objective aka Spock, when we are better off treating it as if it is a salesman.

Your neighborhood brick and mortar, or an online one may care about that while a faceless trans-national big tech will not even register that you left.

Demand will not stop in the next 20 years.

These people WILL fall victim of their own ignorance and upload all their private information to an AI chatbot.

Didn't they literally try this and it failed miserably?

"Alexa but better" is not a profitable business. Amazon profits from Alexa being in people's homes, not from people buying Alexa.

It is a bit comical how the HN crowd spent the last number of decades trying to convince the audience that a programmer's value was actually in their system thinking, not their arcane knowledge of a particular language, but now that the language barrier is removed it turns out it was just the arcane language knowledge after all.

I'd posit that the old school HNers were the ones actually right, though. From what I see in the average Joes is that, while they can communicate with AI, they don't really see how to leverage it.

This has to be build and takes time.

But this is the major thing were ai will be introduced into every single layer of software we have.

The HR System which might have needed x people? The agent has solved 90% of tickets already before they hit a human. Agents will do the repetitive but not that easy to automate tasks.

How is that even possible? I thought they already put AI chats into everything. Maybe AI chat inside an AI chat.

The agent will be running 24/7 and be triggered by different things. Chat input, scheduler, actions someone executes inside the software etc.

The SEO campaign tool? You might need to fill out a lot of different things, configure stuff then execute. In the future you will start the SEO campaign tool, the agent will ask you missing information, it then fills out everything, gets a budget and runs it. It will adjust the campaign based on feedback from the SEO tools.

Kind of a workflow engine but smart, so smart that it will handle A LOT more than what an traditional workflow engine could ever do.

Local LLM usage is bouncing across major runtimes, custom runtimes and forks plus custom quants and variations of open models. llama-swap and some other tools integrate those and provide metrics but as the field expands can't keep up even with AI support.

I need to look at tailscales dashboard/etc again. I started on that when it launched but couldn't get my runtimes to go through it. I think the easiest tie in for metrics/etc is grafana + Prometheus but you need that stood up, and then a custom parsing job for each runtime based on metrics it reports and processing that data into useful dash.

I would love to see more standards here like OpenAI endpoints with metrics and information to link into. ./metrics is great but not standardized.

What makes you so sure? Some would call it "automated running scissors" and want tool firmly in responsible hands. And could assess "automated (indeterministic) reliability" as still far away in time.

Are there any contexts we may have missed that would surely, safely replace individual deliberate action with "the ANN can do it"?

Of course spending less is alluring: but you have to be able to afford spending less. Also considering that giving jobs to people who deserve it prevents other general "boomerang" social costs.

It is really, really easy to use, and I think triggers similar dopamine response that has cause FB so become so prevalent.

I mean, yeah, there are plenty of apps where AI fucking sucks and no one will use it, but I think basic ChatGPT-type use is extremely common already