AI megathread

Inference Is Unlikely to Ever Be a Low Marginal Cost Operational Node, & the Other Reasons Why the Anthropic and OpenAI IPOs Ought to Fail

Brad DeLong

Digital Gods, real costs: why a rational world would see the doom of the foundation‑model-builder IPO, because the AI labs are highly unlikely to ever get profits, let alone hyperprofits. Inference never becomes sufficiently cheap, AI-entity judgment stays bad, and durable quasi-rents flow to NVIDIA & company—not to the model‑makers…

From Paolo Perrone I get four things:

(1) “Inference” is very unlikely to ever become a low marginal-cost node in the system...

(2) Language models now have sufficient verbal fluency. What they do not have is judgment as to which pieces of the human information corpus that has made up their training data are knowledge as opposed to simply s***posting. Hence anything they produce that is not for the immediate assessment by a skeptical human for whom it is part of their information diet requires IMMENSE “babysitting” not to run off the rails...

(3) That is not going to change. Hence belief that these systems are on a trajectory to become human or superhuman in the next decade or so is simply crazy. What works are applications that are modestly right-sized in their expectations...

(4) Plus ones in which the model itself generates bare verbal fluency, and in which the substance is tied down by the ground truth of a scrubbed trusted, organized data store...

What is the big—and bitter—lesson this time? It is this: data cleaning is most of it, as it so often is. And when cleaning the data properly is not most of it, finding the right data is.

Inference Is Unlikely to Ever Be a Low Marginal Cost Operational Node, & the Other Reasons Why the Anthropic and OpenAI IPOs Ought to Fail

Brad DeLong

Indeed... as the below video demonstrates, companies like OpenAI and Anthropic have racked up so much debt that the investors will never see even a 2x return. Ed Zitron has commented extensively on OpenAI and Anthropic's crazy financials and how they are have been in increasingly worsening negative cashflow positions, the worst of which was this year thus far.

He looks at the failed SpaceX IPO, and how SoftBank tried to get a few billion for a loan, which the banks denied, even though the loan is a fraction of SoftBank's current stock market value, due to OpenAI's extremely shaky market position:

OpenAI considers IPO delay as tech stocks plummet | Ed Zitron

“So we’re sitting on this series of time bombs and the fact the people with the money haven’t been asking the questions is genuinely dangerous.”

Author of Where Your Ed At and host of the Better Offline podcast Ed Zitron joins The Tech Report’s Isaac Pound to talk about OpenAI delaying their IPO and the multi-trillion dollar tech stock sell off that has followed a souring in public an investors confidence in AI.

There have been interviews and writings across the years where he's suggested different things but the Platonism in some form or another feels the most consistent.

Interesting. How many Platonist scientists are there these days? Are there any good articles exploring Penrose and Platonism?

Yeah I think the structure of the brain is of great importance, even while not being sure what metaphysics we should consider the "right one". 

I've actually said a few times on here that I do think humans will make some kind of synthetic life ("androids") it just won't have the structure of a Turing Machine.

I agree... structure seems to be part of the picture, but I think that the other half is that the structure seems to have certain requirements that souls find interesting and useful as a vehicle according to criteria we have not even the beginnings of an inkling of.

I think that the brain is less interesting than the whole of a biological organism, as we must also consider plants, fungi, single-celled organisms, and more, that do not have brains, yet display a form of consciousness, sentience, awareness, that is rather alien to our own.

The brain just so happens to be the major organ correlated the strongest with consciousness in animals. It is unclear what the corresponding organ, location or other is for non-human, non-animal organisms.

Yeah even admissions of guilt seem to turn into marketing opportunities -> "Hallucinations".

Anything to brush the major issues and failures of these models under the rug by pretending that they're just another sign of "intelligence" or "consciousness" in these algorithms. When there's billions and trillions on the line if these models collapse in one way or another... well, this just makes the inevitable blast radius all the worse, in size and potency. This wasn't the great reset I was thinking would happen...

The challenge will be that paranormal evidence is either based on meta-analysis from the lab or collections of witness accounts. Personally I am convinced of Survival but I can understand why those who've never had any "weird" experiences might doubt.

I do agree that reality is deeper than what you call the "shallow physical surface".

I have to wonder if that is why some become so... infatuated with these algorithms, falling in love with them, marrying them, and other such projections. Jungian psychoanalysis has been a great help in understanding how and why people become... lost in them.

It is my strengthening connection to the spirits I work with that perhaps offer a protection against such dangers ~ I have clear examples of something intensely felt that isn't physical, and whose structure is animated and conscious just like my own. The animating presence and awareness of their subjectivity I can sense is far more real than the objective structure they inhabit. I can sense, but not perceive their subjectivity in the sense of an object ~ but my imagination interprets that as being object-like, an "orb" representing their central awareness, as it were.

I do think Michael Levin has made some interesting arguments about the Platonic Space and how different structures could instantiate a conscious agent...and how this *might* include LLMs...but I am still unconvinced. Biological structures seem to be directly about instantiating a conscious or at least living being whereas AFAICTell the structure on which the LLM runs could just as easily be running some other programs...

And there are good reasons to reject the Computationalist views of Consciousness.

My shamanic experience of various entities has led me to believe and think that biological structures do not instantiate ~ they rather act as an avatar, a focal point, a container, for consciousness to express itself through. It appears to require a strong and powerful possession of the structures, which then shape, alters, mediates, the possessing beingness, awareness, existence. The structures themselves cannot be conscious or intelligent ~ the possessing being is what animates the structures, and expresses intelligent through them.

No computer algorithm has ever expressed any behaviour outside of the algorithmic constraints, beyond cosmic rays and such causing bitflips in memory or storage ~ random variance makes for algorithmic output that isn't as deterministic, yet that is just still part of the algorithmic constraint.

Early on, we did have experiments in programs that were designed to rewrite the originating binary on-disk, but even that is just more algorithm, with all the same base constraints.

Biological organisms do not behave like algorithms, despite the many flimsy attempts at comparisons. If I examine my own behaviour and thinking, I find... habits, patterns, emotions, beliefs, that shape my choices and decisions.

Materialists may claim that the brain chooses, even if we're not aware of it, but there is no scientific evidence of such things, only the ideological narrative that they parrot mindlessly.

Inference Is Unlikely to Ever Be a Low Marginal Cost Operational Node, & the Other Reasons Why the Anthropic and OpenAI IPOs Ought to Fail

Brad DeLong

Brad Delong also writes

That is clear to me in a way that it was not clear to me back in the day that the Google or the FaceBook or the Microsoft IPOs were unsound. I thought all three of those were very risky, yes. But, even though the valuations seemed very high to me, I did see a possible path to durable hyperprofitability for each.

Right - I wonder how many regrets they who trusted this crackpot back in the day and didn’t invest in Microsoft and Google must have  

Brad Delong also writes


Right - I wonder how many regrets they who trusted this crackpot back in the day and didn’t invest in Microsoft and Google must have  

He said those companies had a path to incredibly profitability that he doesn't see with LLMs, just that their valuations seemed very high.

"But, even though the valuations seemed very high to me, I did see a possible path to durable hyperprofitability for each" 

Interesting. How many Platonist scientists are there these days? Are there any good articles exploring Penrose and Platonism?

His Why Are We Here interview deals with some of this: https://www.whyarewehere.tv/people/roger-penrose/

Also his interview stuff on Closer to Truth: https://closertotruth.com/contributor/ro.../#episodes

[Laird]

The issue isn't that there's no one on the other side, it's that it doesn't seem to matter what we post.

That's because what you post is so biased and one-sided as to exclude and ignore the most important part of the reality. Your selection of biased content is driven by your own demonstrable bias.

As I referenced earlier in the thread, you had asked a skeptical question in a separate thread prior to the public release of ChatGPT, seeming to think the answer was an obvious "No", yet when I asked it of ChatGPT after that release the answer quite clearly was "Yes".

Your tepid response was to implicitly shift the goalposts: now you expected not just correct answers but profound and concise ones.

You've been biased from the start and have never gotten over it.

I had my own bias at the start: along similar lines of thinking as Gary Marcus, I thought that LLMs as mere word-predictors didn't and couldn't have anything akin to understanding, and that some sort of integration with explicitly symbolic modules would be necessary to achieve that. I also thought that, again, as mere word-predictors, they would be unable to develop a meaningful level of anything akin to logical thought.

It's become clear as I engaged more with both the research and the technology itself, and as the technology has improved, that this is a mistaken view: in virtue of their underlying neural networks, these LLMs quite obviously develop a world model and at least a simulation of understanding (taking real understanding to entail consciousness), and they similarly are quite clearly capable of quite sophisticated logical thinking, or, at least, again, a simulation thereof (similar corresponding clarification) - to the extent that the distinction is even meaningful in this context.

So, I've dropped that bias. Unfortunately, it doesn't seem that you're likely to ever drop yours, and to recognise how groundbreaking and revolutionary what's going on is - and, yes, too, imperfect, but to focus on that imperfection as you and those like Gary do (at least insofar as you leverage his output to support your own bias) is to totally miss the point.

Not only that, but you seem intent on playing dumb as to why. "My metaphysical beliefs? Well, gee whiz, I just don't know what on earth they could possibly have to do with any of this. Really, I have no idea what you're talking about."

Uh huh. 🙄

It's similar with self-driving tech. Your focus is entirely on the imperfections, not on the remarkable achievements so far.

To see, directly, a "black swan" with regards to Psi is to have a personal experience. Quite different than using a tool one finds impressive.

Your choice of a minimising framing that overstates the distinction is deliberate, of course: your bias hard at work.

Using a tool is of course a personal experience, and when the tool accomplishes something that you and Valmar say is impossible or not what it seems to be when it quite clearly is, then that is as much a black swan in this context as is a personal experience of psi that skeptics say is impossible in the parapsychological context.

Epiphenomenalism is about decision making. LLMs don't make decisions.

Epiphenomenalism is about more than just decision-making, but in any case, LLMs at the very least simulate human-like decisions, with the same clarification that a "real" decision in this framing would be a conscious (mindful) one, and the same caveat that in this functional context (refer: my earlier functional definitional of intelligence), the distinction may not even be meaningful. That's of course where the relevance to epiphenomenalism becomes meaningful: if a functional decision can be unconscious and mindless, then consciousness and mind could hypothetically be a mere ineffectual tack-on to decision-making.

I think we should be intellectually honest about this, even though there are other good and decisive arguments against epiphenomenalism. I thought that that was the point of Psience Quest: questing for truth in the context of the scientific study of psi and survival. I thought that the serious members here, like yourself, were committed to honesty in the pursuit of that truth. That's one reason why your failure to acknowledge and correct your bias on this topic has disappointed me so much.

Interesting how disappointment can run in both directions.

Your post is a lot of insinuations and apparently includes parts where you've telepathically read my mind...

[Laird]

Oh, well, that lays everything to rest. I'm glad you've cleared it up. There's no bias here to see after all.

[Laird]

Yesterday, Claude Fable 5 was released and, judging from the early reactions, it appears to be another significant step forward.

It's worth documenting in this thread the post-release controversy:

Class of AI Models Hyped as Scarily Powerful Apparently Scared the Government Too Much and Now They’re Disabled
Mike Pearl
June 13, 2026
Gizmodo

Brian Merchant over at Blood in the Machine described it like this:

After sparking a major news cycle in the tech media with its April announcement that it had built an AI model, Mythos, so powerful, so dangerous that it threatened to upend the entire civilizational order—and that it was diligently withholding the product from the public so as to protect us from it—the nation’s now-#1 AI startup decided to put Mythos up for sale after all.

Hours after Merchant wrote those words, the export control directive was delivered to Anthropic, and Fable 5 and Mythos 5 were made inaccessible due to apparent national security concerns. It appears Anthropic was only ordered to revoke access for users who are not U.S. nationals, but it’s understandable that Anthropic would find it impractical to let anyone access them anywhere in the world for fear of disobeying the order. Among many issues, non-U.S. nationals work at Anthropic. It’s clearly simpler to just pull the models entirely until the situation is resolved.

El on her House of El - AI channel has also some some spirited commentary on this saga:

The Claude Shutdown Is a Total Sh*tshow