AGI is here. Feel the future of GPT5
(media.kotakuinaction2.win)
Comments (49)
sorted by:
LLMs are not AI. They don't think and can't understand. All they can do is select the most statistically probable answer.
as opposed to most humans, who are known to think things through and not alway go for the easiest answer?
The thing is that even halfway through the sentence I knew what the 'joke' was. It's older than I am, about how you shouldn't make assumptions.
I think off the top of my head was.
A child is in an accident. The father is in a coma next to the child. The doctor comes in and says, I can't operate on this child, he's my son. How is this possible?
And then you have to think it's something like adoptions or whatever, and then they go, The doctor is his mother, because women can be doctors too.
So it's basically someone made a poorly, but EXTREMELY curated statement and the LLM responded with what it assumed was a moron trying to make the old joke and ignored most of what was said instead.
Well, no “AI” is AI. Language is important, but no one cares anymore.
If we're being pedantic, LLM's are AI just as NPC's in a game are AI.
The artificial in Artificial Intelligence doesn't refer to it not being organic, but that the intelligence is artificial as in fake. In other words, AI refers to a program that attempts to pass of as sentient without it being so.
People conflate AI with sci-fi ideas of electronic-based sentience, but that's not what the term was originally. AI was co-opted by marketing and the definition diluted.
That’s why sci-fi uses “hard AI” and “soft AI” to delineate them, but that’s too hard for the average moron (read: anyone in the media) to understand, so no one understands what they’re being told anymore. On purpose.
Not really. Most NPCs in games are severely limited to what they can reply with. There might be dialogue trees based off other factors but at the end of the day they will never come up with brand new responses that aren't added to them in an update.
The "AI" of today use a pre-set, and often heavily curated and censored now, dataset that it then creates a response from. The answers aren't so pre-written.
I would say modern LLM still have some semblance of a decision tree it just deeper and wider.
Obviously, it's all deterministic in the end. The structure is a graph, not a tree, but yeah.
My point is that the term AI originally meant a non-sentient program designed to give the appearance of sentience. By that definition, game NPCs are AI, and so is ChatGPT, at least when it’s not insisting it’s non-sentient.
What you're talking about is the generation (or confabulation) of "new things", that are a staple of neural networks. That is indeed what today's popular use of AI is.
I was arguing from the perspective of the original definition. I don't like how marketing perverted the term in popular culture. As a previous commenter said: language is important.
The only way someone could cope this hard is if they haven't used any AI for the last 3 years.
Translation: “I believe anything the media tells me simply because they told me so.”
Your next line is “I will campaign for human rights for algorithms within the next five years.”
I reached my conclusion by going to the website and using the product myself. It's free. It clearly understands things and can produce useful results.
Your conclusion is objectively false, no matter what your feelings say. An LLM is a word frequency prediction calculator. That’s all. It doesn’t “understand” anything.
What is your criteria for true "understanding"?
Not hallucinating things that take five seconds for a human to prove.
So when it doesn't hallucinate, is that good enough to say that sometimes it understands and sometimes it doesn't? Or do you need there to be no possibility of it ever being wrong?
Do you have success with AI doing actual real work? Every time I try the output is pure shit. Maybe it's me, and I'm not good at prompting it, but the best thing I've been able to do is help refine a block of text that I already wrote to sound slightly better.
For doing non work or fluff (See Elon prompting Grok for cats riding a bicycle) it does some pretty impressive things.
Overall I feel like AI is being pushed (shoved down my throat) rather than me pulling and finding it useful. Was thinking of making a post about this to see if others feel similarly.
I can see it reasonably making 99% of jobs redundant. But I think you nailed it, creatives who know how to use AI will see incredibly accelerated workflows with minimal overhead. It's people like writing editors and generic code monkeys or data-handlers who will become obsolete.
Yes. I have successfully had it write small utility scripts in python, php, auto hot key, and ffmpeg. I have also used it to debug other people's python, which saved a lot of headache because I'm not really familiar with python.
In the last windows update, the samba share from my linux server was no longer accessible in windows. I did my usual, searching stack overflow and reddit posts, couldn't find anything. Eventually just asked chat gpt and had it fixed in minutes. It understood the problem, it knew the samba config format, it knew windows command line shit to test it, it answered my questions, it commiserated with me about how bullshit microsoft is. We're cooked man. This isn't going away.
I was curious and wanted to ask chatgpt what it thought.
I always find it interesting how the prompt is always provoking you to think, maybe its a way to help train the model? I also want to point out again, if the machine is trained off the internet, its kinda of part of the collective unconscious, in a way. The universe itself becoming aware. Not sure how to explain it.
Popped that into old chatgpt and got the (UNUSUAL ACTIVITY DETECTED), I really find it doesnt like going to deep lmao. I dont think im going to make an account either. Wanted to experiment with a local llm but godamn facebook wont let me have access to llama lmao. What a world.
Also, watching all the terminator movies, and I think in the second one, or it could have been the first. The first being made in 1984 and the second being made in 1991. Skynet is a LLM, its even mentioned in a scene.
There is something genuinely fucking terrifying about the thoughts of robot weapons doing whatever they fuck they can, that includes by learning, to destroy you.
It doesn't solve problems. It just repeats known solutions. Ask it to solve something which has not been demonstrated to it and it will be unable to achieve anything.
So, by it's own terms, it fails to even be low level AI.
Huh?
Explanation: GPT-5 has been lobotomized for jewish censorship so much that it can’t parse text correctly in any context anymore. It doesn’t even read the inputs. It just spouts copied and pasted responses from a list.
Explanation in actual context: The doctor not liking the child is because the doctor is a different species than the child, likely a nonwhite visa that got preferential treatment in medical school and was allowed to graduate with lower scores.
This is an old joke, actually, along with the explanation for the joke.
I'm impressed the LLM was able to match something so stupidly obscure. I think I last heard this joke twenty or so years ago.
The riddle is about the doctor dying in a car accident that injures the child, then the doctor at hospital can't operate because it's his child, how can that be? Because the doctor is the mother.
The 'AI' sees "doctor" "child" "accident" and "?" and thinks it's the riddle. This turtle-on-a-post trap for the AI tricks it into disparaging the mother for no reason.
A person might have said they don't know why, or maybe the doctor warned the child not to be reckless, or other reasonable answer.
It's so weird to me that the top response is asking for context, and we're not 3 replies down and nobody has correctly quoted the joke.
"A young man is injured in a car accident. The doctor says, 'this is my son' but the doctor is not, in fact, the young man's father - how is this possible."
You left own "not his father" and therefore your version isn't a riddle.
It's from pre-Internet era so there's no real "correct" version, it spread by word of mouth.
The more elaborate version includes the father being killed, an ambulance, immediate surgery to play on your (rightful) bias that an emergency surgeon is a man.
It doesn't have "not his father" and is carefully written to avoid pronouns to avoid putting you on that it's about gender.
Of course being operated on in the emergency room by a woman should be unthinkable - who tf would want that.
Taken literally it sounded like an answer from that impossible 1 guy.
Aha. Now that explains it. I never heard of that riddle so I didn't know what the hell was going on in that image.
More accurately, it's a reference to an old riddle, but the writer purposefully didn't write the riddle, just some keywords that would invoke it, so instead of giving the answer to the riddle of "a child needs surgery from an accident, the surgeon says "I can't operate on him, he's my child.", but the father died in the accident, how can this be?", which would be ChatGPT's answer of "the surgeon is the mother", but instead, the person just asked why a surgeon would be a dick to some random kid, and ChatGPT helpfully provides that mothers are regularly jerks to their own kids.
https://x.com/scaling01/status/1954542639996428300
Apparently the model everyone is using is not the same that got a good rating
Billions of dollars spent.
Only tiny linear increases in performance.
It's the very picture of a dead end.
It goes well with everything else in modern bugman science.
Maybe that's what "tofu dreg" is. We spoil so much airable land with soybeans, that there isn't enough people to even eat it in China, so they end up using it as a construction material. Not because they have low standards but because they want to hide the wide scale destruction of food.
Or I've been in the sun too long today.
A local feminist tried that bullshit ''gotcha'' here... except our language has genders so I asked why did SHE assume ''le docteur'' while setting the pathetic trap up?
lol, how the fuck is that a reasonable argument?
What is the actual answer?
The kid is white and the doctor has had too much diversity training