AI computers aren’t selling because users don’t care
AI computers aren’t selling because users don’t care
pivot-to-ai.com
AI computers aren’t selling because users don’t care
301 comments
-
Maybe I'm just getting old, but I honestly can't think of any practical use case for AI in my day-to-day routine.
ML algorithms are just fancy statistics machines, and to that end, I can see plenty of research and industry applications where large datasets need to be assessed (weather, medicine, ...) with human oversight.
But for me in my day to day?
I don't need a statistics bot making decisions for me at work, because if it was that easy I wouldn't be getting paid to do it.
I don't need a giant calculator telling me when to eat or sleep or what game to play.
I don't need a Roomba with a graphics card automatically replying to my text messages.
Handing over my entire life's data just so a ML algorithm might be able to tell me what that one website I visited 3 years ago that sold kangaroo testicles was isn't a filing system. There's nothing I care about losing enough to go the effort of setting up copilot, but not enough to just, you know, bookmark it, or save it with a clear enough file name.
Long rant, but really, what does copilot actually do for me?
-
Our boss all but ordered us to have IT set this shit up on our PCs. So far I've been stalling, but I don't know how long I can keep doing it.
-
Tell your boss you talked to legal and they caution that all copilot data is potentially discoverable.
-
Set it up. People have to find out by themselves.
-
-
same here, i mostly dont even use it on the phone. my bro is into it thought, thinking ai generate dpicture is good.
-
It's a fun party trick for like a second, but at no point today did I need a picture of a goat in a sweater smoking three cigarettes while playing tic-tac-toe with a llama dressed as the Dalai Lama.
-
It's great if you want to do a kids party invitation or something like that
-
That wasn't that hard to do in the first place, and certainly isn't worth the drinking water to cool whatever computer made that calculation for you.
-
-
-
-
Before ChatGPT was invented, everyone kind of liked how you could type in "bird" into Google Photos, and it would show you some of your photos that had birds.
-
The only feature that actually seems useful for on-device AI is voice to text that doesn't need an Internet connection.
-
As someone who hates orally dictating my thoughts, that's a no from me dawg, but I can kinda understand the appeal (though I'll note offline TTS has been around for like a decade pre-AI)
-
longer: dragon dictate and similar go back to the mid 90s (and I bet the research goes back slightly earlier, not gonna check now)
similar for TTS
-
-
-
-
I use it to speed up my work.
For example, I can give it a database schema and ask it for what I need to achieve and most of the time it will throw out a pretty good approximation or even get it right on the first go, depending on complexity and how well I phrase the request. I could write these myself, of course, but not in 2 seconds.
Same with text formatting, for example. I regularly need to format long strings in specific ways, adding brackets and changing upper/lower capitalization. It does it in a second, and really well.
Then there's just convenience things. At what date and time will something end if it starts in two weeks and takes 400h to do? There's tools for that, or I could figure it out myself, but I mean the AI is just there and does it in a sec...
-
it’s really embarrassing when the promptfans come here to brag about how they’re using the technology that’s burning the earth and it’s just basic editor shit they never learned. and then you watch these fuckers “work” and it’s miserably slow cause they’re prompting the piece of shit model in English, waiting for the cloud service to burn enough methane to generate a response, correcting the output and re-prompting, all to do the same task that’s just a fucking key combo.
Same with text formatting, for example. I regularly need to format long strings in specific ways, adding brackets and changing upper/lower capitalization. It does it in a second, and really well.
how in fuck do you work with strings and have this shit not be muscle memory or an editor macro? oh yeah, by giving the fuck up.
-
(100% natural rant)
I can change a whole fucking sentence to FUCKING UPPERCASE by just pressing
vf.gUin fucking vim with a fraction of the amount of the energy that's enough to run a fucking marathon, which in turn, only need to consume a fraction of the energy the fucking AI cloud cluster uses to spit out the same shit. The comparison is like a ping pong ball to the Earth, then to the fucking sun!Alright, bros, listen up. All these great tasks you claim AI does it faster and better, I can write up a script or something to do it even faster and better. Fucking A! This surge of high when you use AI comes from you not knowing how to do it or if even it's possible. You!
You prompt bros are blasting shit tons of energy just to achieve the same quality of work, if not worse, in a much fucking longer time.
And somehow these executives claim AI improves fucking productivity‽
-
exactly. in Doom Emacs (and an appropriately configured vim), you can surround the word under the cursor with brackets with
ysiw]where the last character is the bracket you want. it’s incredibly fast (especially combined with motion commands, you can do these faster than you can think) and very easy to learn, if you know vim.and I think that last bit is where the educational branch of our industry massively fucked up. a good editor that works exactly how you like (and I like the vim command language for realtime control and lisp for configuration) is like an electrician’s screwdriver or another semi-specialized tool. there’s a million things you can do with it, but we don’t teach any of them to programmers. there’s no vim or emacs class, and I’ve seen the quality of your average bootcamp’s vscode material. your average programmer bounces between fad editors depending on what’s being marketed at the time, and right now LLMs are it. learning to use your tools is considered a snobby elitist thing, but it really shouldn’t be — I’d gladly trade all of my freshman CS classes for a couple semesters learning how to make vim and emacs sing and dance.
and now we’re trapped in this industry where our professionals never learned to use a screwdriver properly, so instead they bring their nephew to test for live voltage by licking the wires. and when you tell them to stop electrocuting their nephew and get the fuck out of your house, they get this faraway look in their eyes and start mumbling about how you’re just jealous that their nephew is going to become god first, because of course it’s also a weirdo cult underneath it all, that’s what happens when you vilify the concept of knowing fuck all about anything.
-
The only things I've seen it do better than I could manage with a script or in Vim are things that require natural language comprehension. Like, "here's an email forwarded to an app, find anything that sounds like a deadline" or "given this job description, come up with a reasonable title summary for the page it shows up on".... But even then those are small things that could be entirely omitted from the functionality of an app without any trouble on the user. And there's also the hallucinations and being super wrong sometimes.
The whole thing is a mess
-
-
-
presumably everyone who has to work with you spits in your coffee/tea, too?
-
adding brackets and changing upper/lower capitalization
I have used a system wide service in macOS for that for decades by now.
-
changing upper/lower capitalization
That's literally a built-in VSCode command my dude, it does it in milliseconds and doesn't require switching a window or even a conscious thought from you
-
Gotta be real, LLMs for queries makes me uneasy. We're already in a place where data modeling isn't as common and people don't put indexes or relationships between tables (and some tools didn't really support those either), they might be alright at describing tables (Databricks has it baked in for better or worse for example, it's usually pretty good at a quick summary of what a table is for), throwing an LLM on that doesn't really inspire confidence.
If your data model is highly normalised, with fks everywhere, good naming and well documented, yeah totally I could see that helping, but if that's the case you already have good governance practices (which all ML tools benefit from AFAIK). Without that, I'm totally dreading the queries, people already are totally capable of generating stuff that gives DBAs a headache, simple cases yeah maybe, but complex queries idk I'm not sold.
Data understanding is part of the job anyhow, that's largely conceptual which maybe LLMs could work as an extension for, but I really wouldn't trust it to generate full on queries in most of the environments I've seen, data is overwhelmingly super messy and orgs don't love putting effort towards governance.
-
I've done some work on natural language to SQL, both with older (like Bert) and current LLMs. It can do alright if there is a good schema and reasonable column names, but otherwise it can break down pretty quickly.
Thats before you get into the fact that SQL dialects are a really big issue for LLMs to begin with. They all looks so similar I've found it common for them to switch between them without warning.
-
Yeah I can totally understand that, Genie is databricks' one and apparently it's surprisingly decent at that, but it has access to a governance platform that traces column lineage on top of whatever descriptions and other metadata you give it, was pretty surprised with the accuracy in some of its auto generated descriptions though.
-
Yeah, the more data you have around the database the better, but that's always been the issue with data governance - you need to stay on top of that or things start to degrade quickly.
When the governance is good, the LLM may be able to keep up, but will you know when things start to slip?
-
what in the utter fuck is this post
-
-
-
-
-
The first two examples I really like since you're able to verify them easily before using them, but for the math one, how to you know it gave you the right answer?
-
they don’t verify any of it
-
-
I use it to parse log files, compare logs from successful and failed requests and that sort of stuff.
-
and now we’re up to inaccurate, stochastic
diff. fucking marvelous.-
Stay tuned for inaccurate, stochastic
ls.
-
-
-
-
How about real-time subtitles on movies in any language you want that are always synced?
VLC is working on that with the use of LLMs
-
I tried feeding Japanese audio to an LLM to generate English subs and it started translating silence and music as requests to donate to anime fansubbers.
No, really. Fansubbed anime would put their donation message over the intro music or when there wasn't any speech to sub and the LLM learned that.
-
All according to k-AI-kaku!
-
-
We've had speech to text since the 90s. Current iterations have improved, like most technology has improved since the 90s. But, no, I wouldn't buy a new computer with glaring privacy concerns for real time subtitles in movies.
-
You're thinking too small. AI could automatically dub the entire movie while mimicking the actors voice while simultaneously moving their lips and mouth to form the words correctly.
It would just take your daily home power usage to do a single 2hr movie.
-
-
They're great for document management. You can let it build indices, locally on your machine with no internet connection. Then when you want to find things you can ask it in human terms. I've got a few gb of documents and finding things is a bitch - I'm actually waiting on the miniforums a1 pro whatever the fuck to be released with an option to buy it without windows (because fuck m$) to do exactly this for our home documents.
-
a local search engine but shitty, stochastic, and needs way too much compute for “a few gb of documents”, got it, thanks for chiming in
-
Offline indexing has been working just fine for me for years. I don't think I've ever needed to search for something esoteric like "the report with the blue header and the photo of 3 goats having an orgy", if I really can't remember the file name, or what it's associated with in my filing system, I can still remember some key words from the text.
Better indexing / automatic tagging of my photos could be nice, but that's a rare occurrence, not a "I NEED a button for this POS on my keyboard and also want it always listening to everything I do" kind of situation
-
I wish that offline indexing and archiving were normalized and more accessible, because it’s a fucking amazing thing to have
-
-
-
Apparently it's useful for extraction of information out of a text to a format you specify. A Friend is using it to extract transactions out of 500 year old texts. However to get rid of hallucinations the temperature reds to be 0. So the only way is to self host.
-
Setting the temperature to 0 doesn't get rid of hallucinations.
It might slightly increase accuracy, but it's still going to go wrong.
-
Well, LLMs are capable (but hallucinant) and cost an absolute fuckton of energy. There have been purpose trained efficient ML models that we've used for years. Document Understanding and Computer Vision are great, just don't use a LLM for them.
-
-
301 comments