Maybe quantum computing will be the next leap.
Exactly
You can try but what phones do is more than just matrix maths you can punsh in excel. Id wagor if you wanted to do that (although you certainly theoretically could) you would die of old age before you are done. There are computer vision algorythms for object detection. But even for something as simple as a circle detection the algorythms are complex and more importantly very resource heavy. Which is why people use machine learning to do that. You can read up on that. It certainly isnĀ“t self aware. But it is what we currently get sold as artificial intelligence. Not only as consumers also as developers.
That is so funny yet so true. Iāve never thought of this.
Some people are underestimating what can be done with AI and overestimating what AI is.
Yeah, youāre probably right, and yeah AI and neural networks have been used to death as buzz words. I remember when AMD talked up a new branch predictor because of it being based on a neural net which to be fair it was, but the neural net which runs your branch predictor operates on the order of nanoseconds. There simply isnāt time to be doing anything super fancy there. More complicated networks for predicting weather, modeling drug interactions, natural language processing, or doing image recognition in a self driving car are on another level altogether.
But, the fact that most of what a neural network is actually doing is just dot products and other matrix math doesnāt mean you canāt get complex behavior out of it. Nobody who worked on GPT3 fully understands exactly whatās going on in there. Understanding the inner workings of large neural networks is a whole other subfield of AI on its own.
Very true
Walking bags of self perpetuating chemical reactions became self aware. Took a long ass time, but it still happened.
Thereās nothing to say that the same thing canāt happen with a shit ton of little logic gates.
When you do trial & error by hand, you are an idiot.
When you spend 10 times as long having a computer do trial & error, you are an AI-Engineer.
I see this as the end product, after the heavy lifting has already been done. the machine just needed to be trained to do a specific thing, repeatedly, really fast. it doesnāt need to learn different things from thereā¦
Doesnāt need to start from scratch?
I probably conflate AI and ML, but in my head, the application that uses / benefits from the Machine Learning, is a form of (dumb/dumbed down) AI. it is just a specialist in one field, not general intelligence, and being trained for one taskā¦
that would be my cue to leave
Haha for me I look at more like breaking down to the most simplest sense how the human brain works.
Brain receives information, brain processes information, brain sends out instruction to perform action.
Humans are born knowing nothing but with āpre configuredā hardware so to speak. We learn as we progress through life.
In a very simple sense a computer is not hugely different. It start off knowing nothing but as it receives information and instruction it ālearnsā how to react to various inputs.
Thatās the āgeneralā part of AGI that everything so far is missing. Thereās already plenty of AIs that have superhuman performance in playing Chess, or Go, or some other specific thing, but if you ask the same AI to make coffee, or do laundry theyād fall to pieces.
Making coffee is surprisingly difficult for some humans also which is not really encouraging.
Yeah it is, but you can put a lot of those āspecialists in one fieldā together and eventually youĀ“d get something pretty convincing.
Github copilot is one such things that is pretty impressive. It uses GPT-3 (some gigantic language model) and trains that on all the code on github. So, it can auto complete a hangman game for you including drawing hangman in the console and what not. Want some random data structur? Call your class that copilot does the rest for you.
GPT-3 is as far as I am aware is not available unless you pay for it.
There is this site here that lets you play with an older version of GPT.
https://textsynth.com/playground.html
You can put some text in and it auto completes it with something hopefully convincing that makes any sence at all.
We might get there someday soon where the entier amazon support team can be replaced with āAIā and nobody will notice anything changed. At least the chat support. I donĀ“t know if we can make it talk yet in a realistic way lul.
When you think about it, making a ācup of coffeeā is a difficult job.
In case you ever have a slow month at work, get a notebook and write down all the steps to do this. Be mindful not to overfill any containers in the process
i have few illusions about āaiā on silicon, they are correlation and compute engines.
there is no thought behind anything they do beyond the parameters its given.
that might change once we figure out quantum computing.
What is thought though? how would you define the action of thinking?
I dunno manā¦ the amazon chat bots seem pretty good to me already language-wise. Itās just an n=1 thing, and Iām probably not the best test case since I already know how to talk to a bot, but itās always been pretty painless for me to get to the point where the bot is like āok, let me get a real personā so they can track down a missing package or something.
I think the biggest road block to bots taking over even more customer service type duties are decisions based on business rules. Those often involve classifying intent of the person theyāre talking to which is still pretty difficult sometimes for a bot to get right.
What I ment we will probably see at some point is. What if you never noticed you even talked to a bot in the first place and therefore never even felt compelled to ātalk to a real personā. You thought you where writing with a real person for all this time and it solved the problem with your missing package.
Ofc, I do not have an ETA for this, but I would not be surprised if I live the day to experience it.
Yeah thatās what I mean. Itās probably already happening with some people. Iām just not one of them. The only time I pull up the chat window at all is when I know the bot probably isnāt going to help, and I do need a real person to help fix something, but most people arenāt aware of those boundaries in the same way.