|
Data Graham posted:As much as I roll my eyes at anyone touting AI-anything, I will 1000% believe *GPT or whatever could tell me "what's traffic like up ahead?" which to this day Siri can't make heads or tails of. Siri is a horrible personal assistant. I just use it to try to respond to texts sent to Carplay while I’m driving and half the time it’s a garbled mess. I will say that I don’t know what technology Spotify is using for their DJ thing, but that impresses me by how much it actually sounds like a radio DJ. It plays music you have and that you like and announces all their names with about 98% accuracy.
|
#
?
Dec 3, 2023 16:56
|
|
|
|
| # ? May 20, 2024 05:27 |
|
|
"Where is the nearest Walmart/Michael's/Publix/etc" "Here is a list of <x>, the nearest one is Y miles away" "OK, directions to the first one on the list" "I'm sorry, what? Hi my name is Siri, nice to meet you, I have no memory of any prior interactions"
|
|
#
?
Dec 3, 2023 16:57
|
|
|
"Where is the nearest gas station up ahead?" "The nearest gas station is 5 miles away *directions to gas 5 miles behind on the freeway because it has no concept of the fact that I am driving in a direction*"
|
|
#
?
Dec 3, 2023 17:00
|
|
|
“Add apples to my shopping list” Who is speaking? “It’s me!” … OK, I’ve added nappies to your shopping list.
|
|
#
?
Dec 3, 2023 17:00
|
|
|
I swear this used to work just fine, but now: "Hey Siri, play my Favorites Mix shuffled" "Okay, now playing 'Favorite' by Neko Case."
|
|
#
?
Dec 3, 2023 17:20
|
|
|
statistical models for analyzing and transforming digital data have shown pretty solidly over the last 15 years that they can beat hand-engineered features and rules for things like detecting an object in a picture, translating between languages, converting speech to text, and otherwise mapping between domains the only practical way to generate those statistical models, so far as the modern computing industry has found, is to have them be iteratively refined through machine learning. a team of humans could build the same enormous set of floating point numbers, eventually, but not on human-civilization time scales. importantly, improving these models doesn’t require us to develop a deeper fundamental understanding of the relationship between japanese-in-Kanji and spoken-English-with-a-Texan-accent, it just requires us to be able to say “this mapping is better than that mapping” and feed that fact (ideally also the magnitude of the difference) back in to adjust the model the kinds of statistical models that we know how to build are very good at mapping between domains: finding isomorphic patterns in multiple types of data that share some equivalencies (two human languages, spoken audio and textual representation—and, welcome LLMs, English expressions of facts and queries over those facts). deepl doesn’t “know” (ignoring the metaphysics of what “knowing” is) that the article it’s translating about the war in Ukraine is sad, or propaganda, or written in the style of an epic poem. it just knows how to convert—probabilistically—between Russian and Italian. it works at a higher level than old word-by-word or phrase-by-phrase pseudo-transliteration simply because it has a vastly larger set of equivalencies (and, importantly, almost-equivalencies) to work from than those older systems did. most sophisticated rules are encoded in the huge matrices of numbers than we were practically able to express in the “human-written” rules of before, and because computers got a fuckton faster at this work when Karpathy and whoever unlocked using GPUs to attack Imagenet problems, they can just look at more data at a time. if you give a human translator a paragraph to work with instead of making them treat each sentence independently, they’ll do a better job too! this is why all the gen AI hype bros get wet in their boxers every time someone talks about a version of GPT with a higher token limit there are lots of things that AI could usefully do for us, even in areas that have traditionally required incredible amounts of human investment in training, like medical diagnoses, if we built those systems correctly. doctors spend medical school and then their subsequent experience building up mappings between symptoms and underlying causes, and between underlying causes and treatments, and between unknowns and ways to fill in the gaps. anyone who has ever tried to have a conversation about adult ADHD with a physician over the age of 50 will know that their training can lead them astray too the big problem with AI these days is that people are acting like we’ve already done that work and figured out how to map “symptom space” into “floating point number space” or whatever, or even that we have the techniques and computing power to do that work in the next small-N years. we know a lot about language-related mappings, and humans have long been susceptible to “speaks fluently” impersonating “has a deep understanding”. (ask the parent of a child who started talking early how often they would forget the shallowness of understanding behind some of the sophisticated word combinations or vocabulary.) we know very little about ways to map “state of the road around me” effectively to numbers, or numbers to “actions that this car should take”, and I don’t think it’s clear at all that our success in one domain points to the inevitability of our success in another the other big problem with AI these days is that it works in that it can dramatically reduce the amount of human effort required to produce a result, within increasingly acceptable bounds of quality. artists aren’t worried because StableDiffusion produces obviously-bad images, they’re worried—and should be—because the images that it produces images that are increasingly good enough for some consumers of those images. this represents an increase of “acceptably-good image-production productivity” for society but, uh, you don’t have to spend long looking into who benefits from increases in productivity to worry about that outcome even if you don’t care that your desk sculpture is 3D-printed vs carved by hand (or the 2D equivalent) so for the Siri traffic thing they need to build a bunch of high-enough-quality mappings: speech to text, text to “intent” (give traffic report for <route> over <time period>), traffic data to …some model of whatever dimensions of traffic state, traffic state to English words, words to audio speech. I think we have the technology to do all of those pretty well at this point, though I’m not very well-informed about traffic modelling and prediction; Google generally estimates my travel time pretty well so if all I wanted was “3-5 min delay” or “you’re 18 minutes from your exit, 5 more than normal” then we’re at the “stitching poo poo together” stage and not the “understand the problem well enough to start modelling it” stage. if I want “there’s a Jays game tonight and your dinner reservation is at 6pm so you need to leave work early” sorts of traffic modelling, it’s farther out but attainable if someone wanted to throw the scientist and hardware time at it I mostly think Apple will be the last big company to do anything really ambitious here, because they are so much about (when at their best) the marginal quality of the experience. that’s why they restrict the options of their systems so much: that’s the only way to make 99.99% of the possibilities work, whereas Samsung is pretty happy if you get to 80% of them and if the Pixel 35 interprets “traffic stopped for a long train” the same as “traffic stopped for a recent bridge outage” sometimes then whatever, people will keep drinking that garbage (and if you thought that artists were upset about the encroachment, wait until these things get a little bit better at producing source code or diagnosing program failure. gonna need a dedicated subforum for collecting all those tech tears) E: that looked a lot smaller on my phone, sorry
|
|
#
?
Dec 3, 2023 17:31
|
|
|
Last Chance posted:I swear this used to work just fine, but now: Yeah last week I told Siri to play my discovery station This week she will only play discovery by daft punk I have no fuckin clue what changed but I said the same thing in both cases.
|
|
#
?
Dec 3, 2023 17:33
|
|
|
Arrival estimates are eerily accurate, like it's able to nail my arrival 17 hours later on a road trip without any knowledge of how fast I'm likely to drive or what traffic I will run into, within like 10 minutes. Repeatedly. It feels like the issues with Siri are at the "parsing human language"/"understand intent" level, and I accept that it could be my fault for not talking enough like a computer, but it's a whole nother issue that I feel really dumb giving voice commands and I keep having to reword things and misspeak and try to start over and get all tangled up in knots trying to make it time out so I can try again and not crash at the same time. Star Trek this isn't
|
|
#
?
Dec 3, 2023 17:35
|
|
|
Time to dig up the source code for The Guild of Thieves (1987) text parser and hook it in. Or maybe SHRDLU (1968) 
|
|
#
?
Dec 3, 2023 19:04
|
|
|
What did that guy just post? 
|
|
#
?
Dec 3, 2023 19:44
|
|
|
For me, nothing has topped one of my best friends’ attempts to have Siri update an appointment years ago:
|
|
#
?
Dec 3, 2023 20:19
|
|
|
I feel like I'm very late realising this, but is there no connection between any of the first-party Mac apps with tag support? It would be really handy to have some sort of interface that lets me see where I've used a specific tag in Notes / Reminders / Finder (are there any others?) but as far as I can tell they're all siloed off?
|
|
#
?
Dec 3, 2023 21:57
|
|
|
nexxai posted:I'm sure this is a me problem but I figured I'll ask in case I'm not the only one. One of my favorite sports radio personalities has a saying: "If we're going to hold them accountable for their losses, we need to hold them accountable for their wins." I know we poo poo on Apple for stuff, but in the last macOS update that came out this week, this issue seems to have finally been fixed. I didn't even realize it was better until this evening when I disconnected from the docking station, brought my laptop upstairs, waited for it to do its stupid fucken reboot and it just never came. I had submitted the report every loving time it happened, which was basically every working day between now and whatever day Sonoma was released, so I'm going to tell myself that they got so sick of seeing that identical report showing up that they assigned some dev to fix it specifically.
|
|
#
?
Dec 4, 2023 06:25
|
|
|
I have just upgraded from a macbook 2017 to a macbook pro 2023 (it was a lovely birthday/christmas present.) One of the apps I use all the time is called Xee3 (but the three is the little cubed symbol) and it is an imaging program with a feature I cannot seem to find in any alternative whatsoever - folder organisation. Xee allows me to set up to ten seperate directories and then assign a copy shortcut and a move shortcut for each directory. So for example, I was going through a set of photos to do with some car and location research and I have set directory one to the number 1 and two to 2 and the period is the delete option. I can then simply hit my hotkeys and start moving all my clutter around with ease. Problem is Xee does not work too well on the Mchips I have discovered and it is a roulette as to whether the pictures will load or not. I have been doing my best to find an image program that has this simple and powerful folder management option but no luck. Has anyone here used anything similar? Should I be looking for a different program altogether and not image viewing software?
|
|
#
?
Dec 4, 2023 11:00
|
|
|
down1nit posted:Isn't this a feature called Power Nap? I remember it being a bit buggy at release. A 2007 would be capped at El Capitan 11.6 something. Are you running the latest for your machine? You can start a HAUS thread if you like too. Hey thanks. Yeah it was a typo. I use Spark for mail and don't have any devices connected. I have never spilled any liquids on it. pmset result. code:
|
|
#
?
Dec 4, 2023 13:10
|
|
|
Anybody have experience with SheepShaver or Basilisk II on Apple Silicon? I know the former has working JIT for ARM64, and have heard really good things about performance, but it’s always good to get corroboration from people who know.
|
|
#
?
Dec 4, 2023 15:29
|
|
|
Subjunctive posted:Yeah, I thought so: Thanks for checking, but I probably didn't explain it well enough. You used to be able to take an MBOX export and just point the Thunderbird "local folder" to that directory and read it like a normal inbox. That "local folder" doesn't exist at all on modern Thunderbird as far as I can tell. Maybe I'm missing something glaringly obvious but other sources online seem to agree. i.e. https://documentation.its.umich.edu/mbox-thunderbird quote:Important: We recently discovered that the most current versions of Mozilla Thunderbird are not compatible with the "ImportExportTool NG" add-on, which is required for importing Google Takeout .MBOX files. For this reason, you must download and install an older version (Version 102.9.1) before following the steps in this documentation. Instructions for installing the older version can be found below. https://www.howtogeek.com/709718/how-to-open-an-mbox-file-in-mozilla-thunderbird/ ImportExportTool NG was updated since to work in modern Thunderbird, but it won't import mbox files, and the local folder functionality is gone.
|
|
#
?
Dec 4, 2023 16:07
|
|
|
VG posted:One of the apps I use all the time is called Xee3 (but the three is the little cubed symbol) and it is an imaging program with a feature I cannot seem to find in any alternative whatsoever - folder organisation. Xee allows me to set up to ten seperate directories and then assign a copy shortcut and a move shortcut for each directory. I dearly miss Xee as well, and my experience so far is that you'll probably want to look at a different category altogether. I had hard enough a time to replace Xee just as an image viewer. I ended up with XnViewMP because it had a reasonably competent multi-level folder-diving capability for slide shows and batch conversions, and while it has some sorting capabilities, it's all automated and based on existing tags or meta-data. It's nowhere near as interactive and on-the-fly as Xee would allow. For what you're speaking of, it sounds like you need an equally capable general file manager, but it also needs to support image previews to some useful degree, and then you can use some other tool for the browsing and other baseline manipulation.
|
|
#
?
Dec 4, 2023 18:11
|
|
|
nexxai posted:One of my favorite sports radio personalities has a saying: "If we're going to hold them accountable for their losses, we need to hold them accountable for their wins." Goddammit, it turns out that it just does it the first time after closing the lid instead now. GET UR poo poo TOGETHER APPLE 
|
|
#
?
Dec 4, 2023 20:18
|
|
|
Tippis posted:I dearly miss Xee as well, and my experience so far is that you'll probably want to look at a different category altogether. I had hard enough a time to replace Xee just as an image viewer. I ended up with XnViewMP because it had a reasonably competent multi-level folder-diving capability for slide shows and batch conversions, and while it has some sorting capabilities, it's all automated and based on existing tags or meta-data. It's nowhere near as interactive and on-the-fly as Xee would allow. Oh goodness no  If I knew how to code, or even how to start I would make a program myself. If I wanted to remake Xee what would I need to learn? Is this a question for a different thread?
|
|
#
?
Dec 4, 2023 20:46
|
|
|
VG posted:Oh goodness no I think your best bet is Allusion but if you have the ability to move everything into iCloud Photos for organization, even temporarily, there are iOS apps such as Slidebox that are super useful for this purpose (slidebox can start to chug when you have too many categories to sort into however)
|
|
#
?
Dec 4, 2023 21:15
|
|
|
ApolloOne is excellent but expensive. Edit: But it doesn't do folder management. Sorry. bobfather fucked around with this message at 22:32 on Dec 4, 2023 |
|
#
?
Dec 4, 2023 22:27
|
|
|
VG posted:Oh goodness no Swift and SwiftUI to start building a modern Mac app, Swift (or Obj-C) and AppKit to maintain and update a legacy one. The Unarchiver who make Xee are still an active company making their unarchiving app (The Unarchiver). I wonder if they have an update in the works, or if they'd consider open sourcing it
|
|
#
?
Dec 4, 2023 22:40
|
|
|
Couldn't hurt to drop them an email. Edit: Just got an email back, show yourself coward Warbird fucked around with this message at 00:10 on Dec 5, 2023 |
|
#
?
Dec 4, 2023 23:19
|
|
|
edit: nm solved
actionjackson fucked around with this message at 05:34 on Dec 5, 2023 |
|
#
?
Dec 5, 2023 04:18
|
|
|
Data Graham posted:It feels like the issues with Siri are at the "parsing human language"/"understand intent" level, and I accept that it could be my fault for not talking enough like a computer, but it's a whole nother issue that I feel really dumb giving voice commands and I keep having to reword things and misspeak and try to start over and get all tangled up in knots trying to make it time out so I can try again and not crash at the same time. Star Trek this isn't None of these systems understand human intent in any way you can relate to as a human being. This is a crude and inexact analogy, but *GPT is a family of word arrangers built on picking the statistically best next word, where the notions of "best" and/or "most likely" are built by training on a huge corpus of source texts. Calling it "AI" is a mistake. They've trained in enough responsiveness that it can chatbot you into believing there's more there, but really there's no guiding intelligence. No mental model of the world. It's super-Eliza, not a nascent artificial intelligence. Sort of like a parrot that's been exposed to billions of bytes of things that it can regurgitate, but I'd actually credit a parrot with more self-awareness than *GPT. eightysixed posted:What did that guy just post? A lot of really insightful stuff about the current AI hype cycle.
|
|
#
?
Dec 5, 2023 04:29
|
|
|
VG posted:I have just upgraded from a macbook 2017 to a macbook pro 2023 (it was a lovely birthday/christmas present.) Graphic Converter. Lets you set 20 directories (⌘1–0; ^⌘1–0), and you can set individual shortcuts to either move/copy/alias. Backspace'll move your selection to the trash. The setting is in the "Folders" section of the preferences window, which is far better than where I had actually initially found it, buried in the context menu when selecting an image from the browser. They have had AS builds for ages, so I assume it works just as well there as it always has.
|
|
#
?
Dec 5, 2023 06:45
|
|
|
I'm trying to install graphviz and according to them I should be able to just do a terminal command, code:
BUUNNI fucked around with this message at 01:11 on Dec 6, 2023 |
|
#
?
Dec 6, 2023 01:07
|
|
|
BUUNNI posted:I'm trying to install graphviz and according to them I should be able to just do a terminal command, It is as it should be — sudo doesn't echo back or indicate your password input in any way. Just type it in and press enter, and it should go off do its thing.
|
|
#
?
Dec 6, 2023 01:12
|
|
|
BUUNNI posted:I'm trying to install graphviz and according to them I should be able to just do a terminal command, Type the password in and hit enter anyway, the visual indication of typing simply isn’t given there so that it is harder for someone to overlook your password or its length
|
|
#
?
Dec 6, 2023 01:13
|
|
|
Thank you, that worked, but I'm still unable parse what happens after I input my pw, it returns this directory which I don't recognize....
|
|
#
?
Dec 6, 2023 01:35
|
|
|
you forgot the “port”
|
|
#
?
Dec 6, 2023 01:36
|
|
|
Subjunctive posted:you forgot the “port” yikes.... thanks for catching that! 
BUUNNI fucked around with this message at 01:45 on Dec 6, 2023 |
|
#
?
Dec 6, 2023 01:42
|
|
|
That's not a directory, that's the documentation of the command's usage, telling you what options are available. For future reference on commands like this, that's how you know what you typed wrong.
|
|
#
?
Dec 6, 2023 01:43
|
|
|
Data Graham posted:That's not a directory, that's the documentation of the command's usage, telling you what options are available. For future reference on commands like this, that's how you know what you typed wrong. Ah, that makes sense. Thank you guys for your help!
|
|
#
?
Dec 6, 2023 01:45
|
|
|
It bears mentioning that there is a home brew option as well in that would be a bit less painful.
|
|
#
?
Dec 6, 2023 04:13
|
|
|
Thank you all - I emailed the Xee developers just in case. However the belowZenostein posted:Graphic Converter. Lets you set 20 directories (⌘1–0; ^⌘1–0), and you can set individual shortcuts to either move/copy/alias. Backspace'll move your selection to the trash. literally matches what I need it to do perfectly. I have my whole browser list and some hotkeys and I can finally move all my story research junk around with ease. Thank you so much! Going to buy a license immediately 
|
|
#
?
Dec 6, 2023 09:41
|
|
|
Zenostein posted:Graphic Converter. Lets you set 20 directories (⌘1–0; ^⌘1–0), and you can set individual shortcuts to either move/copy/alias. Backspace'll move your selection to the trash. As the other “I miss Xee” guy, let me just say omg yes. This covers pretty much everything, and does it better. 
|
|
#
?
Dec 6, 2023 11:59
|
|
|
Does my heart good to see ancient apps like Graphic Converter (which dates back to the early 90s) still relevant and kicking rear end today 
|
|
#
?
Dec 6, 2023 13:13
|
|
|
|
| # ? May 20, 2024 05:27 |
|
|
Data Graham posted:Does my heart good to see ancient apps like Graphic Converter (which dates back to the early 90s) still relevant and kicking rear end today Yeah, although the only other one I can really think of is bbedit. Still, if you need to do something that seems even vaguely like they can do it, they'll probably do it beautifully. e. Cyberduck's still around, that's probably old enough. Although I can't really recall using ftp for anything except sending stuff to game systems in the last few years. Even old shareware stuff for os9 and below is pretty easy to find/download without ftp these days. Zenostein fucked around with this message at 13:30 on Dec 6, 2023 |
|
#
?
Dec 6, 2023 13:27
|
|