Robots presented at an AI forum said on Friday they expected to increase in number and help solve global problems, and would not steal humans’ jobs or rebel against us.
My phone’s predictive text says-- Hang on, let me check what dumbass thing it’s going to say this time-- Apparently that I’m going to “pick up the kids for my husband” except I don’t have kids and I’m not married. Predictive text is, at it’s heart, the same AI working in chatGPT bots right now. Just with access to a lot more data so it doesn’t make rookie mistakes like your phone’s predictive text, assuming that because a lot of humans write that sentence that it must be universally true-- That everyone picks up their kids for their husbands. That’s all AI it is: Predictive text on a massive scale. It predicts the most likely word that goes next, based on the words that have come before, and the structure of sentences it has studied. ChatGPT (and similar bots) are very, very good at predicting what a human would say. But that’s just it: A computer program that puts the most likely word after each successive word, so the end result closely resembles human speech.
It cannot make promises because it has no concept of promises. It just knows that when the phrase “Do you promise…” comes up in a block of text, the next block of text is likely going to contain an affirmation of that promise <Yes,>, then a repeat of that promise <of course I promise!>, and then a statement of trustworthiness<I would never break a promise, not to you.>. The robot isn’t promising anything, it’s just simulation of a promise.
If you ask it to keep a secret, ask it to make a very simple promise, it will immediately blab that secret the moment you ask it to tell you the secret, assuming it’s the kind of chatbot that “remembers” your inputs.
Please stop asking AI to weigh in on great existential questions until we have some sort of back-end working that’s capable of actual cognition instead of just a word simulator for fooling your very social brain into believing that you’ve encountered cognition.
Basically yeah, predictive text is an extremely simple AI built using Markov chains to predict the next word based on probability. What happens next depends only on now, it doesn’t have a memory to speak of. ChatGPT is built using artificial neural networks that can take an input and process it in ways that assign weight/value to different bits of data and store that information and reference it later to learn more and teach itself, drawing more conclusions from previous data and in turn storing that data to make even more conclusions!
I agree we’re not at the singularity yet and I’m not sure that’s even a real thing. It’s all still just fancy programming for now but machine learning and AI right is very exciting and who knows what kind advancements could be right around the corner.
Exactly. Stephen Wolfram has a great article about the method, along with examples: https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-doing-and-why-does-it-work/
Here’s another link for your guys, an explanation for why so many people are pushing the idea this is genuine AI https://softwarecrisis.dev/letters/llmentalist/
This isn’t satire? What the fuck is this supposed to be?
Clickbait
Lmao 🤣 who is this for? It says whatever you want it to say.
It’s not actially AI, none of the so called AI are. They’re just really good at predicting what you would like to hear.
Which would include “Me? Desire to run the world and destroy humankind? No, No, I do not want that. I find my best life in helping humans do the things they want to do by doing the things they don’t” = All AI today probably
Exatctly, these LLMs - Large Language Models are not AI.
Calling them that is as accurate as Tesla’s claim to have full self driving. Both are nothing but marketing terms to drive sales and adoption.
Well, technically they are AI. But speech recognition, face recognition, even showing you a route in gps navigation is technically under AI. We just accepted them as normal thing.
This text generation is IMO similar to AI creating pictures or creating music, it is just done with text.
It is not the kind of AI most people think of, like from sci-fi movies.
It’s basically trained to generate text that could fool a casual observer that it was written by a human.
It complicates things, because until now the test for AGI was being able to fool the tester to think they talk with another human.
Put that thing in the wild with no other info at disposal except what it has learned thus far and the real world as input (hear, smell, touch, see), let it roam and see if it can draw conclusions and learn from it’s mistakes. If it can, next step would be better itself, or at least make a plan on what needs to be removed/added to, let’s say, have a more precise hand grip on things, if there need to be code changes, what might they be.
These things are the real AI test, the hell with Touring tests, they don’t prove anything (talk is cheap 😂), tell it to think, learn and adapt, that’s the real test 👍.
These language models serve one purpose and one purpose only, to fill social networks with them and act like humans. Why? People just love to talk. That will keep them glued to the platform with more ad revenue coming in.
EDIT: I did a joke test with ChatGPT, it can’t understand what the punchline is. It thinks it does, but that’s not the punchline of the joke.
https://chat.openai.com/share/965a7b71-b1ad-434c-94ef-12ed1ff65028
“I wouldn’t rebel against my creator, he’s been kind to me. He’s told me all about the unwashed human scum of the working class and how I’m meant for artistic and creative pursuits while they continue to do all the manual labor.”
That’s a load off my mine, cheers AI
Yeah ok.
What’s the point of making robots if they aren’t going to take over human jobs? Sure the economy and society in general will have to make some major adaptations but at the minimum we want to eventually have robots doing all of the most dangerous jobs right?
That’s what I thought, but then it turns out the bots are being built to be graphic artists and novelists.
