Current LLM AIs reveal how much of what we consider "intelligence" is really "just" about language — its ability to encode, transmit, and instruct the processing of information. LLMs function as powerful interpreters of human language, ultimately translating it into machine language processed through computational logic. This appearance of intelligence largely stems from their ability to retrieve and statistically transform vast amounts of indexed content in response to human prompts.

But rather than demonstrating that machines are becoming human, this instead reveals how much of what we thought was uniquely human is actually mechanical capability performed by humans.

So what remains that is uniquely human intelligence, beyond the mechanical?
What are we beyond the language we think in and the actions we perform?

------
And what rough beast, its hour come round at last, slouches towards Bethlehem to be born?

(AI haters alert) A bit of discussion with ChatGPT crystallized some ideas and questions I had, supplementing my miniscule knowledge of cryptographic principles:

Threshold cryptography offers powerful protection - but only if distribution is engineered with the same care as the cryptographic math itself. Random distribution isn't just insufficient - it’s often actively dangerous in adversarial settings.

The key issue is the concept of threat pools vs distribution of the necessary T shares in a T of N secret sharing system. If all T necessary shares are reachable within a given threat pool, then your system is weak. It's weak to compromise and weak to denial of service. So the distribution can't be at all truly random if you really want to protect against threats.

Threat pools can be nation-states or organizations - essentially governments, mafia, terrorists, any organized adversary.

I do understand that @WaywardGeek doesn't want perfection to block good. If perfect could exist, then likely it already would. It also sounds like they WANT it to be possible for (good) persuasion to be able to convince system operators to willingly divulge data to "save lives" when that's reasonable. Whatever good persuasion and reasonable need can come down to. I understand the motivation.

It's a laudable ideal, but I feat it quickly breaks down and becomes useless in today's world. As "rule of law" is beginning to do in some previously liberal democracies.

---
And what rough beast, its hour come round at last, slouches towards Bethlehem to be born? - W.B. Yeats

ChatGPT tells me what I'm referring to is the distinction between SSI (Silent Speech Interface) vs BCI (Brain-Computer Interface). There's significant research along both lines. But my preference is strongly toward SSI technologies, primarily because it's less invasive, and more nearly aligned with the limited desirable use.

SSI still has similar privacy concerns about the potential to invade private/internal thought. Again my preference is to keep the interface external so it's easy to turn it off or completely remove it. I can't readily rip out some brain implant if I stop trusting my control over it.

----------
And what rough beast, its hour come round at last,
Slouches towards Bethlehem to be born?

I'm certainly in favor of safe technologies that can assist people with neurological/mobility handicaps. But really, that's what this primarily addresses.

What is happening with subvocalization technology? That seems like much more central to the major use case for human-computer interface. I don't need my mouse to work like a third hand. I have two hands for that already. What I need is for my intended verbalization to be detected by my computer. Without having to speak out loud around other people.

Sure it's conceivable that direct neural interface can some day become capable of "reading" my verbal thoughts. But my understanding of current subvocalization technology is that it doesn't need to be nearly so intrusive. Apparently even when we don't intend to speak out loud, out brain still generates nerve impulses similar to actual speech, and those can be detected externally. They just need to be decoded - which is exactly the kind of thing various AI/ML techniques excel at.

Like others, I don't know what a workable GOOD solution is. But I do know that the current business model is killing society, politics, and the mental health of our children and ourselves. It's a business model of collecting and selling eyeballs, monetizing attention.

That doesn't necessarily create all our social problems, but the ones it doesn't create, it certainly does amplify and exploit.
The current business model needs to die . . .

--
And what rough beast, its hour come round at last,
slouches towards Bethlehem to be born?

But the right will never fall for it . . . Everybody knows llamas come from one of those "brown" countries in South America. Those people that keep trying to sneak into our borders! Heck, he's probably hiding fentanyl in his llamas . . .

Actually it's Zuck sucking up to the new MAGA elite. Because Trump has criticized him for liberal bias. Zuckbook was long claimed by the right to be unfairly censoring right wing lies. (Because if they're right wing lies then they're really just "alternative facts". Remember?) Now Trump is back in power, Zuck is suddenly claiming to be a MAGA-friendly free speech advocate. Sure, Mark. Long as that's where you think the bucks are.

There was no sentence to parse. That was my point. Neither the slashdot nor the bleeping computer article provided any clue.

"saved our team approximately a week's worth of time," . . . A week compared to what? Six months? Two weeks? It sounds as though if they'd just spent a damn week reviewing the code, they'd have found the bugs without Copilot. Why TF didn't they??! Looks like they cared less about finding the vulnerabilities than about notching a win story to promote Copilot. Sigh . . .

All different, or all the same?

"And she's not only merely dead, she's really most sincerely dead."

You can't fix stupid

"It is actually against the law to not accept cash." You keep using that word. I do not think it means what you think it means.

It PROMISES to move SOME stuff out of the cloud. If you can audit and understand its code enough to trust it. But it still uses its own cloud for networked heavy lifting in blockchain-stored "Trusted Execution Environments" (TEE). Assuming again that you trust the TEE code is really what's running your back end services. I might gamble on the locally installed AI by itself. If it's really capable of running offline, without phoning home. I suspect that on-phone AI will have pretty limited capability though.

You mean like Netflix? Following the leads of Amazon Prime and the early offender, Hulu? Greed abounds. Cory Doctorow tagged it. Enshittification