red75prime

Your post reminds me a bit of another conversation: "It's nothing new, we did that with Utah arrays in the 1970s." Skepticism can overshoot into dismissiveness.

A model has internalized and most likely generalized mechinterp research. Yes, it's expected and nothing insane. Just like internalizing and generalizing math research was expected and not insane. The results of the latter are... interesting though.

Hey! A Pinkelbaum can use some nitrogen.

Things change fast nowadays. Microsoft has access to the latest models. Using AI to screen employees is controversial enough to keep it hush-hush. There's no principled reasons that they can't extract useful training data even with the presence of biases.

Ancient bacteria will be surprised to encounter penicillin. Viruses are less resistant to degradation. But, yeah, it's nothing to look forward to, horizontal gene transfer and all.

The first paper (free-tier chatbots) due to fake references, the second paper (a specialized AI-research tool) due to overreliance on preprints.

This reminds me of "Autonomous driving? Ahhhh! There will be bloodbath!"

Then I'd just marvel at your all-encompassing knowledge.

> The subagent is just as intelligent

Isn't it a common practice to use cheaper models to do the grunt work?

My initial comment was tongue in cheek (but transparent enough I think, given the context). So, let's be clear. I questioned whether it's still true that it's people who are doing all the HR work at Microsoft without LLM assistance.

You, probably, talk about something else.

"We" as in "industry in general" or "we" as in "HR at my company"? Although, I'd be more interested in what Microsoft does.

Maybe something has changed between now and then? Say, some kind of a technology that allows to select eligible employees better.

> LLMs do not replicate brain functionality.

Oh! Do you know how does the brain function? Heck, we don't know in detail how LLMs do their thing. The trivial observation that the brain in not a Turing machine is irrelevant if we are talking about functionality and not computational substrate.

But a Turing machine that implements functionality of a target system is the information-theoretical optimal compression of the data produced by the target system. So it stands to reason that LLMs that need to compress vast amounts of data approximate such a machine.

> C++ is Turing complete. Hell, HTML+CSS is Turing complete.

This is absolutely irrelevant to the question of the limits of LLM training. It's a common surface-level patter matching response, though.

> There's a very obvious and immediate one: availability and size of training data. And compute. And time.

Yeah, practical limitations. And practical results like seven hundred solved math problems show that you need, at least, to stop and think.

> What does quantum computation have to do with brains or LLMs?

If you have to ask that, it means that you haven't dug into all this mess deep enough. Basically, BQP computational complexity class is more powerful than P-class. If the brain employs such computations (although we don't know the appropriate quantum algorithms, so it's just wishful thinking mostly), the corresponding Turing machine is impractically large and slow.

> Says the person who just listed a lot of beliefs in marketing bullshit.

Nah. Learn more.

Obviously, error rate of ChatGPT should be measured by comparing to human diagnoses and when it plateaus, we should investigate a possibility that the number is driven mostly by human errors.

Replicating the brain functionality is magic? Or the brain itself is magic, so we need magic to replicate its functionality? Nah, learn more.

An LLM with CoT is Turing-complete, so there's no wall to its capabilities. There's no theorems that establish limits of reinforcement learning either. There's no evidence of quantum computations in the brain.

All you have is a strong belief.

The server doesn't respond right now (2026-10-25T05:30:00Z). What happened?

The wallet is at $2,016.73

> it already works very well.

Peephole optimization that can't be made global due to combinatorial explosion?

Do you know why are you downvoted? "Stochastic parrot" is so out of touch with reality that you are a perfect example of the human gullibility.

When the future has multiple instances of you, it becomes problematic to define what the Born rule means.

For example, "Many Worlds, the Born Rule, and Self-Locating Uncertainty" notices that the rule doesn't mean anything, but it is rational to behave like it does. It's not a very satisfying position, I'd say.

> because they’ve been caught stealing math research

No. They haven't been caught stealing. Tristan Buckmaster publicly expressed surprise that the OpenAI model has used the niche approach he pursued (with AI assistance). 10,000 agents exploring various approaches make it much less surprising.

As far as I understand it's line 1978: `goto saveStateAndReturn;`. Early return that prevents further processing.

> In other words, the number is fake

In other words, the method of assigning CVEs hasn't changed, but the number of CVEs has risen sharply.

I guess it takes a skill, a lot of elbow grease or both to be able to write different persons' voices well. I wouldn't read too much into it. A sign of an amateurish writer, most likely.

I think people tend to idealize societies that never heard about the universal declaration of human rights. BTW, Highland Papua isn't mentioned in the Wikipedia article. And Highland Papua has an interesting economic history. They didn't have storable food or high-yield crops, so it might have contributed to them being a lot of poorly connected warring tribes.

Provably correct programs verified for the hardware they are running on are possible (seL4 is one of a few examples). Their development is very time consuming though. There are caveats, but it's possible, in principle, to fully block certain attack vectors (for example, network access). The attackers then should go a more expensive and dangerous human manipulation or physical access route.

ASI (artificial superintelligence), theoretically, can exploit the ultimate analog nature of digital circuits (something like Plundervolt, but more intricate), but if we'd get to this stage when we need to defend against ASI, we'd have bigger problems.

What it has to do with the "cultural erasure" I quoted specifically? "Who owns the land" is a different problem that is not made simpler by the fact that the land is subject to tribal conflicts.

Which prior knowledge do we have? A group of engineers managed by public figures that show no signs of omnicidality. A large group of people of all walks, which is very hard to vet.

It's less likely to have an omnicidal maniac in a group of 100 people than in a group of a few billion people.

OK. Higher than average and high wages. Some of them owned workshops after all. The historic lesson stays the same: their approach is a dead end and they'll be crushed (nowadays for the national security reasons).

Robobutler flickers shutters on its cameras. "I know plumbing. That would be $200/month."

It's not necessarily a human who found this one. Coding assistants are so widespread that people don't bother mentioning them.