So, what's the future of mankind? 🤔
-
00
-
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
134 -
rojo wrote:
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
But you still don’t see the big problem coming with deep fakes and ‘hallucinations’; as the accuracy and power improves exponentially, and more trust is given, the proportionately small errors could wreck havoc as humans are removed from ‘the loop’.
00 -
rojo wrote:
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
I agree there are big flaws. Compare GPT-4 for GPT-3 in 2020 or, even better, GPT-2 just a few years before that. The pace of improvement is striking. How long will it continue? Even if only a few more years that’s BIG jump.
Remember, computer vision and image generation were basically impossible 10 years ago. This comic is not that old:
Completely incorrect now.
12 -
rojo wrote:
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
When people ask me what I think of this "leader" in the running community - I point to posts like this. There are no black popes. Is that right? Where'd you go to school again?
You weren't fooled by AI. You were fooled by yourself.
36 -
Hi \"Rojo\" wrote:
rojo wrote:
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
When people ask me what I think of this "leader" in the running community - I point to posts like this. There are no black popes. Is that right? Where'd you go to school again?
You weren't fooled by AI. You were fooled by yourself.
Did someone call me?
Or were you referring to Pope Miltiades or Pope Gelasius?And lest we forget, “ a little bit of mercy, makes the worst less cold, and more just.“
10 -
* … makes the world less cold, and more just.”
00 -
Plot twist: Op is an AI bot.
70 -
still bothered wrote:
don't talk to me about "educated" wrote:
Everyone oohing and aahing over AI are people who have never programmed. You are not seeing anything special or creative.
Those text to picture apps, for example, are just some programmer's idea of how to turn text into a picture. The AI creates nothing. Spend enough time at it (won't take long) and you'll suss out all the algorithms.
The more complex the AI, the more specialized and limited its real-world function can be. And the more fragile and glitch-prone.
State your credentials and why we should believe you as an authority on this topic.
I know how to program, that's why.
Everyone who has programmed much knows these things, you don't need to be a professor of it.
If you can program, you should try out some robot battle games. You write the AI for your robot and it battles other people's AIs. You'll go absolutely nuts trying to make your robot do anything right.23 -
AI stinks.
The stupid NFL, NCAA,MLB, NASCAR, FIFA, etc canned their home based media staffs for India, Saudi Arabia, Taiwan based 24/7/365 AI processing.
Nike is transitioning to Adobe AI Web and Adobe AI Firefly an is dumping USATF staff and Citius, Flotrack, Runnerspace volunteers.
Get ready for crappy AI garbage.
40 -
waiting is foreverness wrote:
ai is dead slow now bc a 4090 for example only has 16k cores
it takes dozens of multi-story buildings full of AI hardware racks in Mountain View to house Google Gemini for example
soon we hope to have chips of billions of cores, one dedicated core per neuron, shrinking the above to a single 1U
it works today at Japanese military bases it'll take time to commercialize the technology
Japanese Government won't let it's AI mil chip specs out to commercial manufacturers. No way Rail Gun, Particle Physics, atom level chips, bio intelligence technologies either.
00 -
No Way tomodachi-kun wrote:
waiting is foreverness wrote:
ai is dead slow now bc a 4090 for example only has 16k cores
it takes dozens of multi-story buildings full of AI hardware racks in Mountain View to house Google Gemini for example
soon we hope to have chips of billions of cores, one dedicated core per neuron, shrinking the above to a single 1U
it works today at Japanese military bases it'll take time to commercialize the technology
Japanese Government won't let it's AI mil chip specs out to commercial manufacturers. No way Rail Gun, Particle Physics, atom level chips, bio intelligence technologies eiAther.
As long as we don't get sued for serving Birria Ramen at our Taco Shop I'm happy.
00 -
don\'t talk to me about \"educated\" wrote:
Everyone oohing and aahing over AI are people who have never programmed. You are not seeing anything special or creative.
Those text to picture apps, for example, are just some programmer's idea of how to turn text into a picture. The AI creates nothing. Spend enough time at it (won't take long) and you'll suss out all the algorithms.
I'm not on the AI bandwagon, but this post is incredibly wrong. Suss out the algorithms?
The program is basically giant matrix of trillions of numbers that was evolved over time to give accurate answers. No programmer designed it, and you're never going to figure out everything about it-- it's not just some crummy simple recipe that a person made up.
Why are people upvoting this garbage?
22 -
One thing my education taught me was to keep an open mind and not make assumptions or conclusions. Even some of the staunchest “facts” fall by the wayside as time and science move on. Einstein and relativity arguably are up for grabs, cause/effect is in dubious straits, human “intelligence” has fallen from grace. To conclude that AI can never match human intelligence is probably a moot point, if h.i. Is only different and not as accurate or capable in the long term. Human intelligence comes from a species that sees no problem in fouling its own nest and destroying itself and staking its future on irrational beliefs in gods and superstitions. AI may ultimately be the only route to survival of our kind’s achievements once we have succeeded in destroying our habitat.
37 -
If ChatGPT trained on the LRC message boards, it would teach itself how to troll humans. Every response would get 10/10.
70 -
rojo wrote:
I'm totally unimpressed with AI so far.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
Once again - Gemini is an AI *product*, not all-encompassing AI.
That's like saying "I'm not impressed with cars. I drove a 1992 Fiat Croma and it was slow and uncomfortable"
OpenAI is the clear leader in the space. Try it out and report back.
Also, if you're only using it as "factual look up" tool like Google, you're not grasping what it's really capable of doing.
52 -
don't talk to me about "educated" wrote:
still bothered wrote:
State your credentials and why we should believe you as an authority on this topic.
I know how to program, that's why.
Everyone who has programmed much knows these things, you don't need to be a professor of it.
If you can program, you should try out some robot battle games. You write the AI for your robot and it battles other people's AIs. You'll go absolutely nuts trying to make your robot do anything right.
OK, you're a 110 IQ Excel/Python jockey. How much work have you done on LLMs, neural networks, deep learning research etc?
34 -
Dear AI:
There's a Pick Six carryover tomorrow night at Los Alamitos. Please give me the winning combination.
Thank You.
00 -
I work as a coder. AI is a useful tool, but it makes mistakes all the time, also it's created way more jobs than it's replaced.
There was also a study by a top software company which shows that code quality has dropped hugely since LLMs have become freely available. Mostly from juniors copying a pasting without understanding.
30 -
TrollGPT wrote:
If ChatGPT trained on the LRC message boards, it would teach itself how to troll humans. Every response would get 10/10.
Rojobot, JaminBot, 430MilerBot, JWHBot, GaultBot
00
What People Are Talking About On LetsRun
No top threads at the moment. Check back soon.