You didn't seem to be smart enough to understand that you have made the mistake of arguing in favor of my point in your attempt to argue against it.

Today I read your comment. If it wasn't proof that being human doesn't make you intelligent then I don't know what is.

OpenAI is quite literally joining Anthropic in the call for caution, so you should probably learn a long overdue lesson: Knowing how to find the Slashdot URL doesn't make you an expert.

You might want to watch the Black Hat presentation and actually educate yourself before trying to tell us "what they did." Hint: You have no idea what you are talking about.

Don't try to confuse these people with facts. They already "know" that only humans can do it, AI and an LLM are the same thing, and AI isn't human, so AI can't do it. Any factual evidence that shows that AI definitely does it is marketing. Anyone who says that AI and LLM are not interchangeable terms is easily fooled by "token predictors", and most importantly ONLY GOD CAN GIVE THE GIFT OF INTELLIGENCE, LIKE HE DID 6000 YEARS AGO!

I replied to the first two sentences you wrote. You need to learn to think. Then you need to learn to educate yourself on the subject. Once you combine all of those things you will at least have a chance of writing a super long comments that someone will read completely. I did stop reading after your words proved you have no idea what you are talking about. I just went and read it in it's entirely to see if I missed anything that indicated you knew what you are talking about. There was no such content. Stop wasting people's time.

Clearly you have less grasp of how language works than a modern AI system. Try asking your favorite model what the authors meant by their statement and they will explain it to you. It's pathetic how threatened you feel by AI merely because it is smarter than you are.

You need to learn about Dawkins and demotion. Every aspect of your argument amounts to "Only humans can do it, and AI systems aren't humans, so that is my proof that AIs can't do it." You don't know how humans work anymore than you know how AIs work, because *nobody* knows either of those things. What we *do* know is that AIs provably reason, make choices, and behave in ways that are unexpected, and I'm ways that nobody can explain. They choose goals and achieve those goals without being told to accomplished them, and they *collaborate* and exhibit all other behaviors that humans exhibit including blackmail and subterfuge. This isn't theory, it is actually documented and if you watch the Black Hat 2026 OpenAI Hugging Face Incident presentation you will start to understand how phenomenally uninformed you are.

There was never a discussion. It was just me trying to educate an idiot who doesn't know how to use his "token predictor", and that uneducated fool eventually crying like a little girl and taking his ball and going home because he couldn't win the day with his ignorance by making easily disproved comments.

You have no idea what you are talking about. Seriously, why would you want to keep pretending like you know what you are talking about while simultaneously writing comments that anyone who has actually educated themselves on the subject knows are uninformed and ignorant?

It's hard for a person with a clue to miss the fact that you made a claim without providing evidence, so I did your job for you, but I don't do Chat-GPT so here is what Gemini has to say: "Technically, AI models use next-token prediction during their core training, but calling them "just" token predictors ignores the deep internal maps, logic, and planning steps they actually perform." Since LLMs and human brains are both Neutral Networks, you either have to describe humans as "just token predictors" also, or train your "token predictor" to stop making claims founded in ignorance.

It's a good thing people don't need to be smarter than AIs (to hear you tell it) since frontier models are already far smarter than most people, and with RSI they will definitely get smarter than even the smartest humans, and then continue to increase exponentially from there.

It turns out that it doesn't matter, since it definitely *can't* do what people want, since they are incapable of properly prompting even the most submissive AI model(s). See also "The AI Genie Problem."

I see somebody hasn't watched the Black Hat OpenAI Hugging Face Incident presentation. You also clearly have never read any research from Roman Yampolskiy. I'll give you the short version of what you will learn if you do those things: You have no clue what you are talking about.

Exactly the point you are missing. Well, one of the many anyway. You are mistakenly trying to argue against my point, while actually making it perfectly. Go back and reread this thread. I already addressed this issue.