- cross-posted to:
- [email protected]
- cross-posted to:
- [email protected]
You must log in or register to comment.
Call it “LLM”
I avoid calling it AI because I like to be as specific as possible when referring to software types
I remember reading someone suggest the term “probabilistic automation”, which doesn’t sound bad.
Everything is either a fancy matrix or something that is locally well approximated by a fancy matrix
Oh hey, I literally just made a comment talking about this. Funny coincidence.
Unfortunately the average human doesn’t know what matrices are so the llm looks like magic. Most people i know don’t understand even basic statistics. I guy at work the other day wanted to warn me about the bots escaping containment and how they’re going to destroy us all. I agreed about the destruction part but he refused to listen to anything about how they actually work. No stats for him, he’s an antivaxxer…
My mind is still blown in how words and concepts can be translated into high dimensional vertexes and through the math of their position and direction retain their meanings, and even be changed into the correct alternate versions by doing calculations on them. That doesn’t mean the vectors understand what they represent, but there’s some core level of meaning there. Same with LLMs. No thinking, they screw up stuff that a reasoning being would figure out. Yet there’s something in even the simplest local model that is far more than just autocompleting sentences. That is part of the mechanics, but not nearly enough to describe it.
I turn off all AI functions because they just get in my way and are frustrating when I’m working.
Occasionally I will use it to format or prep a document template.
AI has gotten better. I pasted in about 10 phrases I’d captured in a meeting for a project, and asked to to prep the document with sections etc.
I expected a generic template with some garbage.
It actually inferred a lot from the project notes and built a 5 page project scope, and included things such as testing parameters and thresholds for acceptance. Which were actually good criteria for the product. It also split it into sensible phases with close out criteria and triggers for next phase.
It suggested avenues of adjustment to make the product work along with back up ideas and the risk levelsof each.
Overall it nailed the product plan and phases and had really sensible suggestions, even with such limited input as a few rough notes.
It did better than most of my colleagues, and what I expected would be 2 days of my time researching and typing, was max 4 hours.
About 1/2 hour setup and generate, and the rest was me moving sentences around, or deleting stuff that might not be 100% relevant.
I just wish we had gotten to this state of things in a better way. Maybe it couldn’t have happened, but at least then we wouldn’t know what we missed and still have what we’ve lost.
I just say it’s a really big excel sheet. Text in on one side, run statistics, letters come out on the other side.
deleted by creator
I don’t understand the first part of the meme. Surely if I “don’t want to call it AI” then that is the opposite of sucking up to trump who wants to call it SI.
“It’s not AI because we can explain how it works” slop.
Brains are just fancy matrices too. That’s why the matrices are called “neural networks”.
The whole universe runs on quantum mechanics which is also just fancy matrices as well.
People want to feel more special than that 😀
Escaping “The Matrix” has a mortal connotation.
The field of AI is large and has some embarrassingly simple programs that have absolutely nothing to do with human (or animal) intelligence.
LLMs still belong to the more complex examples. But get marketed by and to people who have never heard of ELIZA or the ELIZA effect. And it shows.
Most journalists aren’t doing their jobs, and politicians ignore all experts for short-term gains.
Yes! I’m sure it’s pedantic, but the term AI was coined during the 1956 Dartmouth Conference. It was meant to refer to the emerging field of computerized “intelligence” or decision making.
Technically the ghosts in Pacman are properly AI, just a really simple algorithm. The pioneering Perceptron, backgammon algorithms, automated control. It’s all AI, it’s just some of it is very primitive.
Whether or not current LLM based AI are"genuinely intelligent" is beside the point.
That’s cool and all, but an LLM just built me a dashboard for visualizing data, following my instructions and asking clarifying questions when it needed my input. I don’t care what you call it, it’s the most impressive new technology and the most powerful software tool I have seen in years.
There absolutely is a real and useful and even groundbreaking technology here, but that conversation is muted by the way CEOs have pushed it on people like it’s the next industrial age, trying to replace and automate whoever possible, and oligarchs have used fear mongering to basically scam the entire fucking country at once and take a trillion in funds to build AI data centers and use exponential amounts of energy just because it can code a little bit better if you throw another 100 billion at it.
I use it and appreciate some of what it can do, and it’s not perfect but it is still very useful. But for fucks sake they could have handled this a little better than dumping so much into they trigger the next economic crisis.
What are your thoughts on the latest math dump? Reminder that we’re still essentially at the beginning.
Pretty interesting and if anything out of it is real, then it’s a significant breakthrough in machine learning. It means we have computer automation to automate research itself, which hopefully should amplify technological progress and not just replace workers, but it’s concerning that we have to consider that impact in the first place.
Someone else said it’s like a “million monkeys hitting a million typewriters”. That’s incredibly dismissive, but I think there’s also a sliver of truth to it.
The randomness and hallucination/error rate is a spectrum here. It’s not 100% random like a million monkeys hitting a million keyboards, but there’s still some hallucinations that lead to big mistakes.
You see the same thing in computer security. There’s a ton of “slop vuln reports”, but now and then, one is real. The fact that generative AI can generate a real one now and then means that it’s actually doing something interesting and powerful, yet that doesn’t mean we’ll necessarily see it hit 100% ever. I think we have to accept that it generates slop by design and someone needs to sift through slop in all various forms.
I think the AI oligarchs are realizing, and the rest of us are slowly seeing, is that you don’t get exponential improvement by adding more datacenters. Maybe sigmoid or logarithmic. And we might not be able to get enough training data to make the AGI everyone dreamed of.
But there might be this middle ground where humans still actually train and work in these fields and use AI to help automate some aspects like today and sift through some slop when applicable and we greatly accelerate technological progress. This middle ground is not what they promised investors, and they’re going to cause a financial crisis when investors realize it’s slowing down to a halt, and they still have to pay employees.
They had to retract three of the findings almost immediately.
That tells me you can’t rely on whatever model they are using and instead they’re more likely just generating a much larger set of (mostly wrong) random results and then trying to filter out the invalid ones.
The dump is large enough that there should probably be some legitimate results in there, but that’s just emulating a million monkeys hammering away at a million keyboards.
Auto-complete is a legitimate valuable technology and I’d buy the argument that LLMs made for a much better auto-complete (in some situations).
No one’s going to think that a better auto-complete is a trillion dollar technology, though, so they instead wrapped it in a chat interface so you get a better ELIZA and can use that to fool people into thinking it’s more than what it is.
I guess you never saw the original apollo program code…
It’s always a dashboard. 😂
For people unfamiliar with the ELIZA Effect, that was discovered in the 1960s with super early chatbot algorithms.
I attempted a presentation at work in the early days of LLM adoption at work and was told “that is counterproductive to this effort.”
I call it Exported Thought. Or Lazy Braining.
I will admit, though, that its image generation gimmick has been entertaining and useful one or two times when I wanted to quickly create a meme in my earlier days of experimenting with it.
You don’t call it AI because you think Trump is a disgusting human being.
I don’t call it AI because I hyperfocus on stupid things and I’m a pedantic nerd.
We are not the same.
I call them “bullshit generators”
I refuse to call it “SI” because there’s no “I” involved. Same goes for “AI” for that matter. It’s a fancy autocomplete, so LLM is appropriate.
Super Autocomplete?
