[Asianometry] Where is the AI Boom Going?
Posted by TwelveSilverSwords@reddit | hardware | View on Reddit | 35 comments
Posted by TwelveSilverSwords@reddit | hardware | View on Reddit | 35 comments
djashjones@reddit
In the same room as my 3D TV and Tamagotchi, lol
Alternative_Ask364@reddit
ML will see lots of applications in research and stuff. And it will be very useful at partially automating administrative tasks. But we are not even remotely close to AI taking down entire industries or revolutionizing the way we live. Shareholders demanded every company invest into AI and eventually chasing returns that mostly aren’t there and the bubble is going to burst eventually.
Strazdas1@reddit
AI has been revolutionizing manufacturing industry long before ChatGPT made AI sexy.
YixinKnew@reddit
What are good examples?
gatornatortater@reddit
I'm not yet convinced that ML is all that handy at administrative tasks. Its tendency to be wrong and to be unable to check itself is risky.
jaaval@reddit
It’s useful with limited tasks. Like read this long document and distill it down to a few sentences. That can actually shave off a lot of work hours.
But I would not use it to write anything I want to publish.
Alternative_Ask364@reddit
I don’t think it can replace workers completely, but as someone working an administrative job, if all my data entry was automated and I just had to double-check and fix any errors that the computer missed, I’d probably do the work of 3-4 people on my own.
A bigger issue is that lots of large corporations have boatloads of systems they use to manage all of their data, and just getting those systems to talk to each other is a Herculean task, and a required one if you want to automate tasks with AI. Corporate bureaucracy in job security I guess.
gatornatortater@reddit
Yep, but those kinds of jobs where you can work out a way where double checking is notably faster than just doing it are definitely a significantly limiting factor.
Largely people are just making themselves look like idiots right now. They use it to make corporatey sounding lists and crap about things they don't really understand and think it makes them look smart, but someone who knows the topic and thinks it looks stupid. "Why are all these redundant items in this list?" "This one entry looks smart, but nobody ever does that cause its pointless" .... kind of things.
I recently got one from my sister regarding a project I was helping her worth. Apparently it was made my some coworker and forwarded on and the original creator didn't know enough about the topic to see the glaring problems with it.
virtualmnemonic@reddit
Not me. I may of only used the 3d feature on my old TV once, but I use LLMs everyday, at least a dozen times.
djashjones@reddit
I'm quietly waiting for a totally local voice assistant box of tricks, then I can ditch Alexa & Siri.
Strazdas1@reddit
Talking to them is the worst part. so inefficient compared to typing.
Not-the-best-name@reddit
A marginally better Google I am feeling lucky
moofunk@reddit
Google can't do "what's it called when..." to find a particular tool that you don't know the name of and you can't even describe properly via keywords.
ChatGPT will gladly provide you with the technical nomenclature for a subject you aren't well versed in, even in your native language, if you just provide a loose description of what you want to do.
Not-the-best-name@reddit
Exactly what I said. A better Google. Not as big of a breakthrough as the internet and maybe bigger than Google when it was first released.
Strazdas1@reddit
hey man no dissing my Tamagotchi, its the best.
Kronod1le@reddit
Still love our 3DTV tbh, passive 3D TVs especially 4K ones are cinema level of good.
lordbaysel@reddit
More like dot com, things are overhyped, things will fall, but survivors will define next decade or two, because it is tech with insane potential that we haven't figured out yet.
COMPUTER1313@reddit
Hey at least it’s more useful than the NFTs.
Phact-Heckler@reddit
Ohh boyy. I had a "friend" (more like university peer whom I probably took one single class with) who I connected with Instagram whose whole account was just trying to upsell NFT and blockchain things.
One day I posted a funny meme mocking NFT in my story as I just found it hilarious and wanted others to also probably laugh. He messaged me within minutes telling me why I posted it and I should take it down and a bunch of other things.
Coming back to the present, his whole instagram page shifted to AI and there is no trace of any NFT or Blockchain anywhere. Made me realise that there are some people whose whole life is to upsell the latest fad there is.
And honestly, good for him. I see him going to Bahamas, Miami, Mexico while I am stuck in my district at a 9-5. If you can take money from the gullible, who am I to judge. You do you.
COMPUTER1313@reddit
As long as he didn't steal money from criminals: https://www.hindustantimes.com/world-news/argentinas-crypto-millionaire-fernando-algaba-received-threats-before-being-found-dismembered-in-suitcase-report-101690692334648.html
Wander715@reddit
You're right and you managed to trigger the AI bros gj
Kinexity@reddit
Lack of imagination is oozing out of your comment.
RemindMe! 5 years
SpitneyBearz@reddit
AI is awesome when it is used for GOOD things... But when it comes to illegal things... I began to afraid a lot. Especially for elderly people.
Lukeforce123@reddit
Voice cloning and live video deepfakes will just keep getting better...
Strazdas1@reddit
better how? Criminal experts already pretty much openly state its impossible to tell AI cloned voice from real person even with their tools.
awayish@reddit
the LLM craze is overblown as there are fundamental limitations there. but it's still a revolutionary moment for shifting focus onto emergent, scale level features. the future is LLM integrated with world knower and big brain models.
old_news_forgotten@reddit
elaborate please
awayish@reddit
physics grounded models with inner world model e.g. some work at anthropic; integrating llm with reasoning ML arcihtectures such as what deepmind is working on.
rddman@reddit
Although creating a world knower might turn out to be a little bit more involved than creating an LLM.
JigglymoobsMWO@reddit
Scientific and math solving AIs are steadily entering "game changing" territory and are beginning to have broad impacts. They are easier to train than LLMs and their benefits are more tangible (e.g. hyperlocal weather prediction using a single GPU instead of a supercomputer). Expect that to be a major emerging trend over the next 5 years.
endlessfield@reddit
I'm actually curious about this, do you have any papers or talks that you'd recommend? Especially the weather prediction ones, I'm curious about the progress on that front.
Thorusss@reddit
https://deepmind.google/discover/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/
theQuandary@reddit
You notice that "up to 10 days" bit?
To be revolutionary, this AI would need to push past 10 days into tin infamous "butterfly effect" territory.
Instead, we get marginal gains in forecasting and basically nothing for sudden dangerous events like tornados where even small increases in prediction still have large impacts on injuries and deaths (average warning is 13-15 minutes, but there are still tornadoes in the US every year that receive zero warning at all).
JigglymoobsMWO@reddit
This is a great one. I also recommend looking up "neural operators"
ResponsibleJudge3172@reddit
Like the one Nvidia demoed during GTC