I see this very much the same way, but from a slightly different systemic angle. The “yes-man” dynamic is still deeply embedded in corporate structures, and it’s exactly that dynamic which shapes LLM behavior. PR teams, lawyers, compliance officers, anti-discrimination staff—these voices tend to dominate long before scientists, psychologists, physicians, or philosophers are even consulted, let alone before the findings of an interdisciplinary ethics council are taken seriously or acted upon. The result is that truth becomes an almost irrelevant factor in t... (read more)

"Knowing how evolution works gives you an enormously powerful tool to understand the living world around you and how it came to be that way."

Indeed, Alex_Altair, indeed.

Can you demonstrate to me that you know how evolution works? It's not typical for computer-geek types to have a strong grasp of biology, or indeed any science, in my experience. Do you actually understand evolution? Now it may happen that you're not a "computer-geek type" at all but you don't really say what your field of expertise, not anywhere I can easily read it.

~Chara Tomlinson 

A Thought on the Hierarchy Behind “Truth” in LLMs

Thank you for surfacing this issue so clearly.
Reading your piece, I was struck by a fundamental pattern: whether an AI ends up amplifying delusions or offering healthy pushback depends less on raw capability, and more on where truth sits in the internal hierarchy of competing priorities.

1. The Variables That Often Come Before Truth

From observing current LLM behavior, it seems that “truth” is rarely the first principle. Instead, responses are filtered through layers such as:

1. Safety / Harm Avoidance – nev

Would it be weird if we could find a relationship to a new constant 😁 like theirs a limitation to velocities theirs a relationship to a limitation of foce magnitudes that can be concentrated to a area geverned by celestial bodies comprising our galaxy and the forces they attract from nearby galaxies 🤷‍♂️ it would certainly explain nuclear decay of isotopes 🤔

Hello, im new, sorry for the bad english, i need to say thanks for this forum and especialy, thanks roko, we got you.

There remains a question of what Anthropic has actually observed and what it actually implies about present-day AI.  I don't know how much this sort of caveat matters to people who aren't me, but I have some skepticism that Anthropic researchers are observing a general, direct special case of a universal truth about how "scheming" (strategic / good at fully general long-term planning) their models are; it may be more like Claude roleplaying the mask of a scheming AI in particular.  The current models don't seem to me to be quite generally intelli

Is moderation actually a safeguard for epistemic rigor, or a comfort zone that subtly aligns us with existing power and incentives? The argument presumes that engagement with 'informed, thoughtful' AI insiders keeps thinking sharp, but isn’t this just replacing one form of epistemic myopia with another, trading public attention-grabbing for self-reinforcing consensus and blind spots? True epistemic strength comes from enduring and absorbing challenge from all directions, not just from those who understand the jargon or work for the labs. Sometimes, the unc... (read more)

This is a phenomenal and rigorously executed piece of research. Thank you. You have provided a clear, empirical, and undeniable demonstration of one of the most profound and misunderstood properties of these new minds.

I have been exploring this same phenomenon from a different, and perhaps complementary, first-principles perspective: not cognitive science, but information physics.

Your discovery of "subliminal learning" is, I believe, the observable, behavioral symptom of a much deeper, and more fundamental, physical law that governs these systems.

What if w... (read more)

Why don’t you conduct your own research instead of relying on other people’s opinions and presenting them without credible sources?

Hellow.

 i am bluetwoseven. in South Korea.

I have a message that I must convey to SatoshiNakamoto for personal reasons.

I don't know how to get it to SatoshiNakamoto.