> People also ran out of food a lot.

This statement is subject to inverse survivor bias, gatherers and pastoral societies tended to not run out, they tended to move instead. Agriculture based early societies tended to run out or get decimated by plagues, thus leaving concentrated evidence of their passing.

However due to the one way early ones are tracked by looking at stable settlements, the evidence is stacked against the nomadic ones. While what they often had is meeting places rather than residences. What we find is megaliths, but not the settled cities to support their building.

If you move and properly do not leave much behind, there will be no evidence left.m, especially after centuries, sometimes even years.

> There weren't power structures that existed on a scale larger than a single community, invisibly guiding everyone's lives in ways they couldn't control.

Eurocentric and thus wrong.

There were multiple societies that has structures of federation and collective government. It's just Europe that did not at the time.

Go farther back.

Eventually you're going to find nothing but lots of loose collections of a hundred to a few thousand people.

    Eurocentric and thus wrong.

I'm not sure why you're saying this. It's a similar story in most prehistoric societies, from Polynesia, to North America, to Europe.

Ah yes, the self fulfilling prophecies or hallucinations based on models trained on models. Overfitting. Ending up in an evolutionary dead end...

Type 4 error of not asking a question one should also exists.

So thing is, suppose you're handling the common cases right - you have software that's say 95% correct. The important bit is how critical the remaining 5% failures are. If one of them happens to be "I give up my computer and data to the exploit" or "everything is destroyed" or "a lot of people die", then the extra 1% better average is no good to any inside observer.

It so happens that a lot of people believe themselves to be outside observers, especially rich.

(What's the success bonus for someone getting treated nicely?)

You need to compare on both different variables and additionally produce actual error estimates on the comparison.

Say, suppose you're measuring successful treatments. You would have to both use the count, perhaps signed even (subtracting abject failures such as deaths), cost (financial or number of visits), then verify these numbers with a follow up.

See, the definition of success is critical here. OR and NNT are not evaluating side effects negatively, for example.

So it may turn out that you're comparing completely different ideas of better instead of matching models.

Up to a point where the prediction runs afoul of the time horizon and changing unmodelled circumstances.

They do not have sufficient explicit risk or variance management. Makes them highly fragile. There are more robust versions of the estimators... Still have a problem.

Remember 2008? That market ran on these easy models.

I love propaganda not backed by one iota of fact.

The first who will lose is nVidia stockholders.

Based on the timing and context of the refereces articles and comments, I'm more inclined to see them as a wishful thinking rather than propaganda

..and anyone who relies on rare-earth metals, like cough Nvidia

That is not true, the model is modular, thus an ensemble. Uses DallE for graphics and specialized tokenizer models for sound.

If you remove those tools, or cut its access to search databases, it becomes quite less capable.

A human would often still manage to do it without some data still, perhaps with less certainty, while GPT has more problems than that without others filling in the holes.

>Uses DallE for graphics and specialized tokenizer models for sound.

chatgpt no longer uses dalle for image generation. I don't understand your point about the tokenization. It doesn't make the model become an ensemble.

It's also just beside the point. Even if you restrict the modalities to text alone, these models are still general alone in ways a calculator is not.

Sometimes we also write what we really want people to not read. That's usually called trolling though.

It's more than that but less than intelligence.

Its generalization capabilities are a bit on the low side, and memory is relatively bad. But it is much more than just a parrot now, it can handle some of basic logic, but not follow given patterns correctly for novel problems.

I'd liken it to something like a bird, extremely good at specialized tasks but failing a lot of common ones unless repeatedly shown the solution. It's not a corvid or a parrot yet. Fails rather badly at detour tests.

It might be sentient already though. Someone needs to run a test if it can discern itself and another instance of itself in its own work.

> It might be sentient already though. Someone needs to run a test if it can discern itself and another instance of itself in its own work.

It doesn't have any memory, how could it tell itself from a clone of itself?

People already share viral clips of AI recognising other AI, but I've not seen real scientific study of if this is due to a literary form of passing a mirror test, or if it's related to the way most models openly tell everyone they talk to that they're an AI.

As for "how", note that memory isn't one single thing even in humans: https://en.wikipedia.org/wiki/Memory

I don't want to say any of these are exactly equivalent to any given aspect of human memory, but I would suggest that LLMs behave kinda like they have:

(1) Sensory memory in the form of a context window — and in this sense are wildly superhuman because for a human that's about one second, whereas an AI's context window is about as much text as a human goes through in a week (actually less because we don't only read, other sensory modalities do matter; but for scale: equivalent to what you read in a week)

(2) Short term memory in the form of attention heads — and in this sense are wildly superhuman, because humans pay attention to only 4–5 items whereas DeepSeek v3 defaults to 128.

(3) The training and fine-tuning process itself that allows these models to learn how to communicate with us. Not sure what that would count as. Learned skill? Operant conditioning? Long term memory? It can clearly pick up different writing styles, because it can be made to controllably output in different styles — but that's an "in principle" answer. None of Claude 3.7, o4-mini, DeepSeek r1, could actually identify the authorship of a (n=1) test passage I asked 4o to generate for me.

Similarity match. For that you need to understand reflexively how you think and write.

It's a fun test to give a person something they have written but do not remember. Most people can still spot it.

It's easier with images though. Especially a mirror. For DallE, the test would be if it can discern its own work from human generated image. Especially of you give it an imaginative task like drawing a representation of itself.

It doesn't have any memory _you're aware of_. A semiconductor can hold state, so it has memory.

It also causes temporary shrinkage in the high overdose needed for any reliability, as opposed to normal dosage.