AI agents are programs that use your logins to do things.
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
@futurebird the problem with technology at the moment is that programmers made a very complicated and expensive algorithm for determining the likely next word in a document (basically fancy autocomplete) and then I guess all programmers got raptured and the marketing team took over?
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
@futurebird I can sort of understand why someone would delegate this task to an LLM because you can have it do more complicated step-by-step tasks in natural language instructions, but even then you'll only get like 80 to 90% accuracy with what you want from it and it'll probably hallucinate something wrong about your task.
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
@futurebird The way they're usually integrated, you run an instance of a service that has the passwords, all in one place, and not ... entirely ... under the control of anyone but you. (It's more complicated than absolute).
Having a single place it all lives makes it relatively easy to maintain.
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
@futurebird as I understand it, one of the factors making IFTTT less useful was that more and more services that you might have wanted to connect it to were taking away their public APIs because they didn't want to be interoperable - they wanted to lock you in to their own systems. An LLM can interact more or less like a human, so it can get around these barriers at the cost of being unreliable in whole new ways.
-
@futurebird The way they're usually integrated, you run an instance of a service that has the passwords, all in one place, and not ... entirely ... under the control of anyone but you. (It's more complicated than absolute).
Having a single place it all lives makes it relatively easy to maintain.
Everything was in one place with IFTTT too?
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
It's still around I see. I wonder who is still using it?
-
It's still around I see. I wonder who is still using it?
@futurebird I know people who still use it. Mostly tiny automations and not something IoT-esque like turning lights off but kinda like sending a notices if some web changes or something.
-
Everything was in one place with IFTTT too?
@futurebird Yeah, but it did so little that at least when I used it, there was always somewhere else.
-
It's still around I see. I wonder who is still using it?
@futurebird I actually have one automation set up for it. It monitors an RSS feed, and then sends a slack message if there’s a post meeting certain conditions. Very simple and effective, and fits into their free usage tier.
-
AI agents are programs that use your logins to do things. For example and AI agent might notice your bank balence is low due to all your polymarket gambling and sign you up for a $890 "financial literacy bootcamp" and send your spouse a text telling them to check in on you. It's the future. But hold on. We've seen this before.
Remember IFTTT?
It also used your logins to do things, but you had to write the programs yourself. I used it to make my lights flash red whenever congress passed a bill.@futurebird Can I rant about the "primitive AI" topic of AI agents being individual members of a botswarm?
-
@futurebird Yeah, but it did so little that at least when I used it, there was always somewhere else.
@futurebird Also my system specifically (I don't run any such thing), I could just plunk the tools down next to home assistant on the same machine and have no passwords involved at all. And it could talk to my network too. No passwords, just local control.
-
IFTTT wasn't a terrible idea. "turn off the lights when I'm more than 1 mile from home" isn't a bad automation. But #IFTTT failed, mostly because it just didn't work reliably. Coordinating the logins and apps was difficult. If you changed a password everything would break.
Why is it better to have an #LLM generate IFTTT task for you? I'm not just asking to be mean I really want to know.
We've done this. What did we learn from IFTTT?
@futurebird I don't think there's anything mean about querying the purpose of inaccurate tools used primarily for convenience.
-
It's still around I see. I wonder who is still using it?
@futurebird I still use it with regularity!
-
@futurebird I know people who still use it. Mostly tiny automations and not something IoT-esque like turning lights off but kinda like sending a notices if some web changes or something.
IIRC people who used IFTTT for post propagation from one service to another usually caught blocks from those who didn't know/didn't trust it. Wonder if that is still a thing.
-
It's still around I see. I wonder who is still using it?
@futurebird There's a lot of people that still use IFTTT in the self-host home automation circle.
-
AI agents are programs that use your logins to do things. For example and AI agent might notice your bank balence is low due to all your polymarket gambling and sign you up for a $890 "financial literacy bootcamp" and send your spouse a text telling them to check in on you. It's the future. But hold on. We've seen this before.
Remember IFTTT?
It also used your logins to do things, but you had to write the programs yourself. I used it to make my lights flash red whenever congress passed a bill.@futurebird No thank you. That sounds dangerous. Some people still write checks and some ATM machines, like at diners, mom and pop stores, etc, don't clear until later or the next day even. What happens when someone writes a check or does a large purchase with their ATM Debit card that doesn't clear right away, nor informs the AI control account, but AI has moved money around, esp lower income with limited income or unsophisticated AI users ticking settings and permissions they know nothing about and it's just not there to cover the charges? Bad checks charges? Theft by deception charges? Will there be an "AI caused it" defense?
-
AI agents are programs that use your logins to do things. For example and AI agent might notice your bank balence is low due to all your polymarket gambling and sign you up for a $890 "financial literacy bootcamp" and send your spouse a text telling them to check in on you. It's the future. But hold on. We've seen this before.
Remember IFTTT?
It also used your logins to do things, but you had to write the programs yourself. I used it to make my lights flash red whenever congress passed a bill.It’s worse. Far worse.
IFTTT can’t “decide” to donate 98% of your savings to a maga pac. An agentic LLM might.
Oh, you set safeguards?
Nice, how do they work? Same way LLMs work generally. Like hiring toddlers to watch your toddlers. Except that toddlers actually comprehend.
-
J jwcph@helvede.net shared this topic
