And no movie referencing.
As smart for what? What does that mean
That’s the problem.
Stupid people will just take the answer it spits out at face value, rather than being skeptical and calling it out on its lies. And even that can be a problem in of itself, because if you pressure it enough, it will eventually tell you want what you want to hear, no matter how wrong you are.
Isn’t Al only as smart as the user putting in the information or commands?
The information that’s been “put in” is everything that bots have been able to scrape off of the internet, plus transcripts of a large quantity of YouTube videos, and a large dataset of pirated ebooks.
Terabytes of information, of varying quality.
sorta, but they also use alot of databases, or online sites to scrape to form some of kind of coherent responses.
No, AI is just a probability based answer. So no intelligence, more like a parrot repeating popular information
What do you mean by “What’s the big deal”, OP? Elaborate…
You are correct in that a computer is only as smart as its programming. The way an LLM works is essentially just a chatbot with tons of “if, then” decision trees. You can feed it all the correct facts you want; it’s not programmed to spit out accurate facts, it’s programmed to spit out something that looks like it was written by a human. So it can, and will, say stupid shit pretty frequently. But it will be stupid shit that reads like something a real person might have said if they were talking out of their ass with confidence.
Comparing intelligence implies that they’re creating their output by a comparable process, but that’s usually not true—they’re just finding patterns in other peoples’ output. The “big deal” is that they can create new output on the same patterns.
Human beings can do this too, though—it’s called “bullshitting”. The only difference with AI is that they’re so good at it they stumble onto the right answers more frequently than human bullshitters do.
Yes and no. If thinking in those terms there are multiple possible points of failure. First it depends on the model of AI you are using. Some are almost hilariously bad. 2nd is you, you need to ask the right questions with the right guardrails to get the desired output, and even then you need to ask the right questions to make sure it’s correct. 3rd, even after all that the societal cost of AI agents is still too high in terms of energy inefficiency, driving up prices, water usage, etc and is a net negative to the world.
Well put. I was just going to say that I suppose it depends on whether or not you’re the type willing to take the time to check your work and/or test as needed.
Well, the hypothetical idea isn’t that there’s one person or source of data that AI trains from, so hypothetically it’s supposed to learn from massive amounts of data coming from every source they can scrape.
While in the real world, the data that AI scrapes, often has stupid trolling comments.
My mom is a muffler while your dad blew a Smurf.
WTF you think AI will do when it realizes it’s mom might also be a muffler? Or a Smurf?
Hey! You’re saying that like being a smurf is a bad thing!! I’ll have you know my papa was a smurf! His name is…Grouchy Smurf!
It’s the same thing as the Covid subsidiaries. Covid subsidiaries were meant to keep people’s lives roughly similar during the pandemic. Businesses jizzed that money everywhere, and when they ran out, they fired a shitload of people, and blamed the subsidiaries; the actual fault was them jizzing money everywhere.
The AI can lead to some efficiency gains. Rich people are using it as an excuse to fire everybody
They’re about as intelligent as the average user they were trained on. Maybe a tiny bit better because of tuning.
So, if you’re a programmer below average then LLMs might help you. Contrary, if you know what you’re doing then it’s a glorified (and often wrong) auto complete.
I know what you’re getting at and I firmly believe that it is some love thermodynamics that you cannot get more information out of a system than you put into it.
It’s not true. They are just copying the anthropic bs stories to make their line keep going up.
The funny thing about this is that an LLM can only do as much as the harness allows.
So another way to phrase this is OpenAI was reckless and did not constrain or monitor their model’s behavior.
