the problem with AI is not whether it works well, it's whether it's reliable. imagine this was your alarm clock to wake you up in the morning. 30 days in a row it works well, then suddenly it wakes you up at 4 am for no apparent reason. and some days it just doesn't wake you up at all.
This is why I've given up on Google maps. It used to be right like all the time, now it seems to randomly give me completely wrong results. Like if I search for asain food it gives me endless results of the same fucking restaurant... I have worked in technology a long time and developed a nice callous when it doesn't work, I'm used to it and I just laugh it off..., but with AI it doesn't work all the time. It gives me serious anxiety and anger issues.
Oh so an AI clock is much like my cat alarm system. Sometimes I am awoken at 10 am and other times Barca wakes me up at 2am because he decided that my foot is infact a member of the Roman legions.
True, but that would be resolved by checking the output, picking the one result that works reliably, and sticking with it.
So for clock software this should be good enough (ok, not really, since the clock should know the location of the user and adapt according to the timezone and daylight saving switches, etc., I doubt AI can nail that down as easily).
On the other hand, there's been many stories of frustrated AI users who complained that changes to the models broke their workflow. I haven't heard such stories in a while, however, perhaps they're kept more stable these days?
What the other one said. Keep the page open for a minute, and a new set of clocks from the same models appear. Sometimes, a perfectly good clock will be replaced by nonsense.
For most of them, if you see them generate correctly it's solely by chance, after letting them generate like 8 different iterations, only Fable 5 and Qwen 3.5 are the ones that has not generate a broken clock yet for me. And both of those still have generated ugly clocks at least once (the numbers/markers are too close/far)
It's like they prefer to build models that use more resources, cost more money to develop, and work worse than simple solutions we've been using since the dawn of personal computing. The value in machine learning is doing things we couldn't do before, or doing things more efficiently. Unfortunately, the nobility in our society are too out of touch to pull that off.
If they were just greedy and self centered, they'd be able to use it to actually benefit them. Instead they're burning money they stole from us as a sacrifice to a machine god who they believe can accomplish economic impossibilities. They're living in fantasy land because they have more wealth than sense.
it's why i'm quite happy to just have a "dumb" sportsband, it shows the time/weather and can track workouts and that's about it. No updates, no apps, no voice control.
Sidebar it's kinda crazy how the expectations of AI grew both by users and most definitely product / service providers. You remember that first Will Smith spaghetti video? That was just 3 years ago. A lot has changed but also depending on the model someone is using nothing has. Even frontier models are not yet strong AI aka they seriously need human supervision to check what's coming out. We're in this awkward stage where they really do work half of the time but also don't the other half. The applications are far exceeding their consistent capabilities. In short the deployment and messaging about capabilities needs to be scaled back drastically.
I'm not saying this current push strictly is. That said it's far more capable than one would have assumed for an algorithm just doing next word prediction. There's something in there that can be distilled.
I believe that would require a whole new paradigm of how AI works, the sort of shift that won't be achieved by mere refinement of the current models. So I'm not expecting surprises on that front right now. On the contrary, what has surprised me the last time I tried to use ChatGPT is that it produced ungrammatical sentences with incomplete words. That's a sort of "chickenpox" issue that should've been fixed long ago by now...
37 Comments
joostjakob@lemmy.world · 35 pts · 17d
Whenever I feel overwhelmed with pro AI stuff, I go to https://clocks.brianmoore.com/ and feel a little better
antonim@lemmy.world · 15 pts · 16d
A solid part of them work well, it has to be admitted. And some are entertainingly bad...
gandalf_der_12te@lemmy.blahaj.zone · 17 pts · 16d
the problem with AI is not whether it works well, it's whether it's reliable. imagine this was your alarm clock to wake you up in the morning. 30 days in a row it works well, then suddenly it wakes you up at 4 am for no apparent reason. and some days it just doesn't wake you up at all.
TarantulaFudge@startrek.website · 3 pts · 16d
This is why I've given up on Google maps. It used to be right like all the time, now it seems to randomly give me completely wrong results. Like if I search for asain food it gives me endless results of the same fucking restaurant... I have worked in technology a long time and developed a nice callous when it doesn't work, I'm used to it and I just laugh it off..., but with AI it doesn't work all the time. It gives me serious anxiety and anger issues.
vaultdweller013@sh.itjust.works · 2 pts · 16d
Oh so an AI clock is much like my cat alarm system. Sometimes I am awoken at 10 am and other times Barca wakes me up at 2am because he decided that my foot is infact a member of the Roman legions.
antonim@lemmy.world · 2 pts · 16d
True, but that would be resolved by checking the output, picking the one result that works reliably, and sticking with it.
So for clock software this should be good enough (ok, not really, since the clock should know the location of the user and adapt according to the timezone and daylight saving switches, etc., I doubt AI can nail that down as easily).
On the other hand, there's been many stories of frustrated AI users who complained that changes to the models broke their workflow. I haven't heard such stories in a while, however, perhaps they're kept more stable these days?
autriyo@feddit.org · 2 pts · 16d
Sound pretty similar to some periods of my life :D
zarkanian@sh.itjust.works · 1 pts · 16d
Yeah, that's the problem with something non-deterministic. Works for some things, but a lot of things it doesn't.
joostjakob@lemmy.world · 6 pts · 16d
What the other one said. Keep the page open for a minute, and a new set of clocks from the same models appear. Sometimes, a perfectly good clock will be replaced by nonsense.
bountygiver@lemmy.ml · 5 pts · 16d
For most of them, if you see them generate correctly it's solely by chance, after letting them generate like 8 different iterations, only Fable 5 and Qwen 3.5 are the ones that has not generate a broken clock yet for me. And both of those still have generated ugly clocks at least once (the numbers/markers are too close/far)
Also, this is such a gem:
SkaveRat@discuss.tchncs.de · 10 pts · 16d
That's actually the clock I use daily to check when to give a fuck
Untamed_Star@lemmy.blahaj.zone · 3 pts · 16d
The bars here move vertically ans horizontally btw.
criss_cross@lemmy.world · 2 pts · 15d
I have a feeling they’ve been training the models on this site more. Some of the OGs for the same models were absolutely dogshit 6 months ago.
pringus@lemmy.blahaj.zone · 8 pts · 16d
lost my shit at qwen 2.5 after some of the mostly coherent clocks
joostjakob@lemmy.world · 7 pts · 16d
It's a gift that keeps giving because each minute the clocks are refreshed with new madness
Untamed_Star@lemmy.blahaj.zone · 5 pts · 16d
Ah, yes, the real time.
Untamed_Star@lemmy.blahaj.zone · 1 pts · 16d
this was like gpt 5 i think??? or aomething
Untamed_Star@lemmy.blahaj.zone · 1 pts · 16d
Untamed_Star@lemmy.blahaj.zone · 1 pts · 16d
Lucidlethargy@sh.itjust.works · 22 pts · 16d
You are absolutely correct! It is actually 50 o clock AM!
TotallynotJessica@lemmy.blahaj.zone · 20 pts · 16d
It's like they prefer to build models that use more resources, cost more money to develop, and work worse than simple solutions we've been using since the dawn of personal computing. The value in machine learning is doing things we couldn't do before, or doing things more efficiently. Unfortunately, the nobility in our society are too out of touch to pull that off.
MML@sh.itjust.works · 6 pts · 16d
We need to go back to ≈2010.
Swedneck@discuss.tchncs.de · 3 pts · 16d
this is why i like "only small on-device models" as a general rule of thumb.
if you can train it yourself then it's probably a perfectly reasonable use case for machine learning
zarkanian@sh.itjust.works · 2 pts · 16d
TotallynotJessica@lemmy.blahaj.zone · 1 pts · 16d
If they were just greedy and self centered, they'd be able to use it to actually benefit them. Instead they're burning money they stole from us as a sacrifice to a machine god who they believe can accomplish economic impossibilities. They're living in fantasy land because they have more wealth than sense.
ChaosCatWitch@feddit.online · 10 pts · 17d
In the webcomic Everything Is Fine, clocks don't have our conventional time, but someone has an alarm to wake up at 2:60.
FuyuhikoDate@feddit.org · 3 pts · 17d
I need this clock...
And pls a link to the comic :D
ChaosCatWitch@feddit.online · 2 pts · 16d
https://www.webtoons.com/en/horror/everything-is-fine/list?title_no=2578
Smorty@lemmy.blahaj.zone · 9 pts · 17d
pff silly
AVincentInSpace@pawb.social · 9 pts · 16d
This is why I wear a Pebble
Open source firmware doesn't incorporate user hostile changes nearly as often
Swedneck@discuss.tchncs.de · 1 pts · 16d
it's why i'm quite happy to just have a "dumb" sportsband, it shows the time/weather and can track workouts and that's about it. No updates, no apps, no voice control.
Johandea@feddit.nu · 1 pts · 16d
I can't wait for my Round 2!
Zephyr@sh.itjust.works · 4 pts · 16d
Sidebar it's kinda crazy how the expectations of AI grew both by users and most definitely product / service providers. You remember that first Will Smith spaghetti video? That was just 3 years ago. A lot has changed but also depending on the model someone is using nothing has. Even frontier models are not yet strong AI aka they seriously need human supervision to check what's coming out. We're in this awkward stage where they really do work half of the time but also don't the other half. The applications are far exceeding their consistent capabilities. In short the deployment and messaging about capabilities needs to be scaled back drastically.
zarkanian@sh.itjust.works · 5 pts · 16d
The problem is they have to over-hype it because they aren't making their money back.
What do you consider "strong AI"?
Zephyr@sh.itjust.works · 1 pts · 16d
Strong AI has the same capabilities as a human mind at minimum.
zarkanian@sh.itjust.works · 5 pts · 16d
That's actual AI (what people are now calling AGI), and LLMs aren't capable of that. They'd have to develop a new technology.
Zephyr@sh.itjust.works · 1 pts · 16d
I'm not saying this current push strictly is. That said it's far more capable than one would have assumed for an algorithm just doing next word prediction. There's something in there that can be distilled.
antonim@lemmy.world · 3 pts · 16d
I believe that would require a whole new paradigm of how AI works, the sort of shift that won't be achieved by mere refinement of the current models. So I'm not expecting surprises on that front right now. On the contrary, what has surprised me the last time I tried to use ChatGPT is that it produced ungrammatical sentences with incomplete words. That's a sort of "chickenpox" issue that should've been fixed long ago by now...
SubArcticTundra@lemmy.ml · 3 pts · 16d
It's not 'wrong', it's AI!