they are quite literally not “intelligent” in any way shape or form. they are large language models that don’t even conceptually understand the words they use, because it is literally impossible for them to do so.
Just tossing in my 2 cents as someone who actually moderates this platform:
My view on generative LLMs is that they are a blight on society so that does bias my views. Putting morals and ethics aside, however, LLM content has been a net negative on this forum and much of it falls under the category of spam or is very borderline at best. Most AI threads have required moderation due to quickly devolving and practically never producing anything even vaguely constructive.
I’d only want an exception for machine translation as that’s the only practical use case I have seen on these forums or any given social media platform.
They are called AI because they are developed using knowledge/techniques from the Computer Science field that has been called AI since the 50s. It’s not much deeper than that.
it’s also called ai because that markets a lot better than llm or diffusion
llm, diffusion = tech jargon for weird machines
ai = wow like skynet in terminator? i must invest in this company at once
I like “Virtual Intelligence” personally
Mass effect universe use that to tell the difference between actual intelligence and a simple imitation.
“Though they appear to be intelligent, they aren’t actually self-aware, just made with clever programming.”
Fits pretty much to a tee to what we have if you ask me.
And yet like a redditor, here you are complaining about vibe rather than criticising their arguments 
As I think more on it, yea it may be preferable to have an exception on LLM posts only for translation. I can’t easily remember any instances of an English speaker using an LLM on this forum and it actually “improving” the text in any way that isn’t just spelling & formatting. In many cases I can remember the output’s been excessively long-winded and as you say, not constructive.
One other useful thing LLMs can help with is making a native English speaker’s text easier to understand for non-native speakers.
For example: (Note this paragraph was generated to be a shitty as possible and the LLM switched genders for some reason in the first little bit)
She decided to bite the bullet and call her mother, but when she finally did, he gave her the cold shoulder and the conversation went south; still, she kept her head above water while juggling three projects, trying not to count her chickens before they hatched. Meanwhile, everyone at the party was spilling the tea and throwing shade about the proposal that had died on the vine, yet some insisted they were on the same page and swore she was the bees’ knees — talk about a storm in a teacup.
If you prompt an LLM with:
can you please help remove the idioms and other language that makes the following hard for a non-native English speaker to understand?
It can help transform it into something that is easier for a non-native english speaker to read.
Ex:
She decided to force herself to call her mother. When she finally did, he ignored her and the conversation became unpleasant. Despite that, she managed to cope while working on three projects and tried not to assume any would succeed. At the same time, people at the party were sharing gossip and criticizing the proposal that had already failed, but some said they agreed with her and praised her highly — it was much less important than people made it seem.
But yeah, LLMs are pretty much worthless. Here’s a screenshot of an LLM hallucinating it’s ability to read a web page but ‘manages’ to summarize it for me in relation to an issue I was having. Just making poop up.
SeaChest tools are not used by many people, they help control power settings on hard disks… The LLM was also making up cli parameters, they sounded reasonable, but didn’t exist. Absolutely a waste of time.
He quite literally is a redditor, with a 10 year old reddit account, that cries about Reddit constantly lol. He does it all the time…on the darktide subreddit.
kill clankers behead clankers roundhouse kick a clanker into the croncrete
Yes, it’s a algorithm. But it “learns” because of the algorithm. By taking what it “learned” and applying it to what you wrote as a prompt it can craft a wholely unique response. Get 10 people to write the exact same prompt and they’ll each get a unique answer.
It’s not general intelligence, it doesn’t “think”, but it’s specific intelligence in that it’s “taught” to do one specific job and “learns” how to do that. In the case of LLMs and other generative AI, that’s responding to written prompts. It’s the same with writing machine-learning code to teach a bot to drive in Track Mania with “rewards” and “punishments”. It learns over millions of iterations what gives it a “carrot” and what gives it a “stick” and does better and better to get more and more carrots and less and less sticks. That’s “intelligence” whether you think it is or not. It can’t do literally anything else, but it’s learned to do that one thing. LLMs learned to reply to user prompts.
Edit: To put it another way. Humans are “programmed” to produce and receive noises. Well, assuming all the hardware and software is functioning properly anyway lol. Anyway, we naturally know how to produce a sound and we naturally receive sounds without being taught, but what we are taught is how to interpret those sounds we produce and receive as language. Similarly, an LLM is programmed on how to produce a response and receive a prompt, but it needs to learn how to do that in a way that’s considered “language.” We’ve only seen the “it’s usable” part of LLMs, but anyone who’s looked into machine learning algorithms would know that when it starts out it’s not capable of producing anything that could remotely be considered a response. If our human ability to learn a language is intelligence, then how is an LLMs, or any other machine learning algorithm’s, ability to learn how to produce a result not intelligence?
this is a really, really stupid take and sorry if that offends you.
Are you suggesting that… you’re on Reddit!?!
I think a simpler way of framing whether or not it’s “intelligent” is not by trying to analyse the system’s percieved intelligence (fraught with anthropomorphism and superiority complexes… just ask cetaceans). We have already left the Turing test far behind, and it turns out it wasn’t that hard.
Instead, and far more objectively, the field of AI is simply interested in replicating human intelligence with machines. LLMs use neural networks in a facsimile of human neurons, because it turns out that gets some pretty interesting results.
All this to say pretty much this:
(except well I do think it’s a little deeper as it has quite an interesting history imo but arguably it’s not)
Wait a second… we are having a discussion about LLMs on a public forum… is this… AI generated content???
Yeah, basically the “is it really AI?” debate is leaning too heavily into the philosophical question of “what is intelligence?” and less a discussion of the capabilities of machine-learning neural networks.
Most forums don’t allow spambots to just post nonsense. Users posting LLM generated nonsense are pretty much the same thing.
yeah removing ai slop would be awesome
