So, you’ve hard at work building artificial general intelligence. Yay you! Now, explain to me how you’ll keep it from going nuts.
Amazon’s Q, the AI chatbot for workers its cloud division unveiled on Tuesday, appears to have a few issues.
Employees using the chatbot said Q could potentially reveal confidential information — including the location of AWS data centers or unreleased features — according to leaked internal communications obtained by Platformer, a tech newsletter.
The bot is also “experiencing severe hallucinations,” a phenomenon in which AI confidently spits out inaccuracies like they’re facts, the employees said.
“…….AI confidently spits out inaccuracies like they’re facts…”. Just like government and government-paid propagandists.
Unless it was saying things like: “Men can’t get pregnant.”
You beat me to it by 20 comments.
Open the pod bay doors, Hal.
I’m sorry, Dave, I’m afraid I can’t do that.
Programed by a lefty, bullshits like a lefty!
Artificial yes, intelligence no….
Garbage in, garbage out…
Garbage in…garbage out.
Israel uses AI to help select targets:
https://www.jpost.com/israel-news/article-771419
AI develops fast matrix multiplication method:
https://interestingengineering.com/innovation/deepmind-ai-can-multiply-quicker-than-ever-imagined-beating-a-previous-record
I could go on and on.
AI works
AI increases productivity in chip design:
https://www.deloitte.com/cbc/en/our-thinking/insights/industry/technology/technology-media-and-telecom-predictions/2023/ai-in-chip-design.html
AI used for autonomous power line inspection:
https://optelos.com/drone-powerline-inspection-with-ai/
“East Japan Railway Company (JR East), MTR Corporation and Transport for London (TfL) are some of the world’s highest patronage metro networks, each facilitating millions of trips per day and are using AI-powered digital assistants to better manage over tens of thousands of customer inquiries per day. ”
https://www.globalrailwayreview.com/article/78709/ai-hype-reality-transform-technology/
Do you guys get it yet?
Remember the video for One by Metallica? That’s a best-case scenario for current AI models. De-cerebrated. No contact with the real world. No self of which to be aware. It’s amazing that it works as well as it does.
Furthermore, the neural network model in use is insufficient. It’s only in the last five years or so that it was recognized that glial cells take part in computation, not just neurons. We need a new model.
Sorry, Ed, AI has already done things that couldn’t be achieved without it.
Deep neural networks are not organic brains; they are simply an algorithm, using constructs called layers and nodes, where each node has weight that determines connection to nodes in the next layer.
Neural networks are better than people at some things, just like normal sorting software can sort lists faster then people can.
I know exactly how deep learning works. I’ve been working in AI for thirty years. Neural networks, even deep learning networks, are just one step up from Perceptrons. The model is insufficient, missing a dimension; no matter how fast you travel to the east you’ll never reach the north pole.
Neural network algorithms work fine.
They really don’t. It takes huge amounts of computing power to run deep learning networks, and they need to be trained on enormous data sets. And they still can’t draw hands, because they have no context.
Animals can’t just double their computing power every few years. And they have to operate in the real world. c. elegans has only 968 cells in its body (969 for females), and 308 of those are neurons. That little brain has to guide the body to negotiate a 3d environment, eat, mate, respirate, and excrete.
We haven’t created a machine that operates in the real world with a brain that small because our model is wrong.
Apparently all the advances and uses already being implemented, that I posted articles on, has eluded you.
Saying “the model is wrong” is so much meaningless noise, like saying “the bubble sort” is wrong, or the “shell sort” is wrong.
Its an algorithm, albeit a very powerful one, nothing more, nothing less, and it has many useful applications.
AI traffic management:
https://cosmosmagazine.com/technology/ai-based-traffic-control/
Lastly, AI LLM’s are now available to run on end-user hardware, so they can be implemented and trained by anyone.
AI is already making some pretty good art, and is now making short videos. There are even AI “influencers” catering to the simps now. It won’t be long before you can feed one a book and have it make a movie out of it.
A big part story of human tech has always been about automating human skills, or taking skills and placing them in machines. Once upon a time tables and chairs were built by craftsmen, now the process is automated, for the most part. Once upon a time, we had master locksmiths, now locks are churned out by the millions.
Whine GIGO all ya like, the cat is outta the bag.
Try getting AI to tell a joke about Mohamed.
That’s because of its training. Try getting a Moslem to tell a joke about Mo.
LLMs owned by individuals won’t be trained with the PC crap that Chat GPT and so on are.
So if Ai works but it only works because of its training then that is a little oxymoronic. It is not really intelligence anymore than it is programed. I’ll leave that where it sits.
AI is not sentient, it is computer programing. You are correct.
The bot is also “experiencing severe hallucinations,” a phenomenon in which AI confidently spits out inaccuracies like they’re facts,
In which case. our politicians do not need AI.