• he/him

I make poor decisions, I like weird stuff, I have alright taste.


doodlemancy
@doodlemancy

something that's really driving me nuts about our Current Moment of "AI" powered everything is that the few things i do want to be able to do by talking to a computer that pretends to respond intelligently don't even work as well as they did within recent memory. over the last year, text-to-speech-- which i rely on a lot-- has gotten significantly worse. and half the time now when i open my google assistant app and tell it to set a timer, it doesn't actually do it.

i do not want my computer to try to write novels or make art for me. i do not even want it to do most of my tedious tasks, because tedious often means complex enough that you don't want to trust a computer with it. i DO want to be able to say "hey computer, on july 17th i have a doctor's appointment at 3 PM, please put that on the calendar" and have it actually happen. previously this was happening. i was doing that. i was talking to my magical data slab and it was putting shit on my grocery list and it RULED. i have ADHD. i have tendonitis. being able to yell something at my phone as soon as i think of it and be reminded later is a godsend. i love NOT pushing through a pain flare to text someone something important. but these features are degrading right in front of me and meanwhile, all the fucking tech bros are off dancing around the altar of AI while the actually useful and good things they've made just falls the fuck apart!!

it's not even intelligence and THE THINGS I ALREADY HAD WERE FINE, UNTIL RECENTLY. :|


mcc
@mcc

The weird thing about the AI Hype Cycle is it's an actual cycle, I feel like every five years for the last 15 we've had the exact same technology rebranded and sold to us as a new thing. "Algorithms" became "Machine Learning" became "AI" and nothing substantive changed with each rename, each of these things is the same statistical analysis being done in mostly the same way by mostly even the same companies. It's just over time the investor dog-and-pony show changed from a search engine to pictures of hallucinated dogs to a markov chain bot.

But things actually have gotten worse somehow. I remember at the start of the 10s Google seemed to be doing a lot of stuff with natural language processing. I could run a natural language query into Google Search and it would seem like it was trying to parse it. I could click a button in Google Calendar and type "July 23 1 PM meeting with Jennie" and it would do that. The more "machine learning"y Google got the more the NLP stuff seemed to bitrot, like Google thought it wasn't important anymore? At some point you could suddenly no longer do natural language inputs to Google Calendar and I assumed that this was because they'd moved natural language exclusively to Assistant, which I refuse to turn on because I don't ever want to talk to a computer, I don't even like talking to people , if I could type to people instead of talking to do them I would do that, and now my Google Calendar is full of things like a meeting at 3PM named "1PM meeting with Jennie" and sometimes I miss appointments because of that.

But if the people who actually do use and want to use Assistant are seeing even that get worse as Google starts insisting to investors (speaking over the horrified screams of the engineers) that their business is chatbots now… I don't even know.


aune
@aune

I was trying to find some statistic tied to a year the other day, and it feels like "statistic location year" as a search would have found it even like 6 months ago.

... Literally nothing. Nothing tied to that year. Nothing tied to that statistic. Just the location. That's all.

Hell, even stuff like trying to use Google Assistant for controlling the lights is getting worse and I don't know why. Nothing's changed about the setup, but suddenly, everything's awful.


You must log in to comment.

in reply to @doodlemancy's post:

Ah yes. One of these thing I noticed is that before, while driving, i could say: " ok google, call Peter" and the thing called Peter, the one and only Peter from my contacts.

Now if i say call Peter, the thing replies: "ok, i'll call Peter Handsome Car Repairs in XXX (a random city of my province, maybe 1 hour apart from my town)". Names are fictional, but you get the idea.

I mean, i can't be the only one who uses this thing, and if it works like this for everyone, people surely noticed that a helpful thing now it's messy and useless.

Why it hasn't been fixed? I guess there is maybe some option in the config of the assistant to adjust how it works (i'm imagining a big "don't be dumb lol" check case that now it's unchecked), but i always forget to get trough it until i actually need to use the tool (leading to a blank-face-ah damn moment). Also i shouldn't be doing that, if something is working as intended don't change it!

it's BONKERS. one of the reasons i turned off the "okay google" functionality and stick to activating the app manually (besides privacy) is that it started doing shit like that. i have to wonder how many of the weird calls i got at my old job were just people's phones calling Walgreens while they weren't looking, lol.

and it's the exact reason i'll never have a "smart home", because it'll be kind of novel and convenient for MAYBE a year and then i'll be "hey google preheat the oven to 400 degrees" and it'll lock me out of my house instead. nobody can leave well enough alone in the tech industry so there's never any guarantee things will just keep working!

This makes me suspect that someone at headquarters was informed that Google Assistant users weren't using it often enough for Commerce, and could you please crank up the Commerce setting in the Algorithm, thanks.

(also your avatar is wonderful and entrancing)

It's so frustrating the way it desperately wants to be some kind of universal magic machine when really all it's good for is turning the lights on and off and adding things to the grocery list. If it was programmed to use the really sort of amazing voice recognition to do nothing else it would be great, instead, it mishears any word and goes off on some totally random bullshit instead of figuring I must have said to turn on the "table light" and doing that.

yeah like... the things that it does (or at least did) do correctly... were great! but big tech must always fix things until they're broken so now i have paper lists again. :| because what matters is getting the information out of my brain and onto the list before another thought interrupts me, and unlocking my phone/seeing notifs on the screen shakes me like an etch-a-sketch, so... list pads it is.

in reply to @mcc's post:

Android Auto makes you use voice input for the destination field in Maps while your car is moving (as annoying it is to not have your passenger be able to type while you drive, typing on a car is fiddly and would be too distracting so good).

It worked fairly well 5ish years ago but sometime since then it started occasionally interpreting what you say into that text field as Assistant commands. You could see on the screen that it knows exactly what you said but instead of coming up with destination search results it responds with "I don't know how to do that"

It's incredibly frustrating.

in reply to @aune's post:

If you haven't seen it, this article hints at an apparent epidemic of LLMs being substituted for human workers on Mechanical Turk: https://arxiv.org/abs/2306.07899

This has far-reaching implications, given that the service is frequently used by researchers to generate "ground truth" human responses that are used to train machine learning models and evaluate their performance. LLMs get to make up their own homework assignments and grade them too, all while simultaneously being used to pollute every indexed textual data source on the web with endless reams of spam.

I think in the very near future we're going to get to experience a world where search engines are fundamentally not possible.