naasking

I'd say that one data point is not enough to object to, except if what's depicted is literally impossible, like the AI generated images of a female pope. Beyond those, the distribution of a set of generated images is what implies likely bias.

> I just don't think that I can ever trust an xAI model knowing that they are actively trying to shape its replies to fit a political narrative.

Maybe you should base this assessment on more than just vibes. Grok came out pretty balanced on independent assessments, where most other models were heavily biased: https://github.com/washingtonpost/political-bias-llm-eval/

> In Europe, criticizing any politician is very very common.

Criticizing politicians has gotten people arrested in Europe. This doesn't happen in the US.

> As for work, in US, research grant were removed for using some "woke" keywords for political/ideological reasons. In Europe, grants are decided by the peers, and when a controversial subject is not attributed a grant, it is not for political/ideological reason, it is because the scientific value is considered as poor.

This is naive. You think European scientists are immune to political influence, ideological conformity and fads?

> I didn't see them defining "for everybody" what low/high morality are.

Morality by definition is a set prescriptions that everyone ought to follow.

> this whole "I am so smart I cannot see right and wrong" is a totally transparent, low-IQ, and low-morality schtick.

The problem with this position is that it takes as a given that morality is in some sense trivial where millennia of debate over it has shown that it is in fact, not at all trivial.

> that they aren't particularly vocal and obviously active about their politics

Is Grok obviously vocal and active about its politics? Or are you talking about Musk?

> I don't see how political commentary about Musk's is somehow forbidden

Nobody is saying it's forbidden, but this is (or was) a technical site, so presumably one would hope that the main topic of discussion is technical.

I don't believe for one second that this metric has been rigourously evaluated, it's also just based on vibes.

> The child knows how to catch the ball, without understanding.

I disagree that decoupling "knowing" from "understanding" is well motivated or useful in any way.

> I don't think that it is useful to say that LLMs understand anything they say, or that we say to them.

This is starting to seem like a Chinese Room disagreement, so I'll just say that what that thought experiment demonstrated is that there is no robust or meaningful way to distinguish semantics from syntactic processing given everything we know about reality, and that LLMs were a fantastic demonstration of this illusion, eg. that "semantics" are reducible to a relational network of syntactic elements.

Not sure why this is weird when you really think about it. You're saying that a multiparameter constraint satisfaction problem is harder to solve, even when there's only a single solution (one point in the total space that satisfies all constraints), than just being given that solution/point. Well of course it is!

> Which simply dooms the terminally ill to fake treatment.

I wish people would stop saying this. First, controls aren't necessarily "fake treatment", they are often compared to other standard treatments.

Second, the treatment being tested can actually harm the patient more, therefore the people receiving your alleged "fake treatment" can actually come out better off. Which is the "fake treatment" now?

I don't disagree with your final point, but mainly with this increasingly pervasive and wrong framing of RCTs.

I think you need understanding to reason, but you don't need reasoning to understand. A child understands how to catch a ball without reasoning about forces, air resistance, gravity, etc.

I think LLMs understand without reasoning. They've built a large associative network of concepts (a kind of understanding), but we don't yet have a good handle on the process of reasoning using that network.

I think it's a mistake to disentangle their abilities from understanding. Just swallow the pill that they have some form of understanding, even if it slightly differs from ours. I really don't see the problem.

1. Determinism is a very small subset of what makes computers useful. Non-determinism like stochasticity is literally everywhere, like random seeds.

2. LLMs are detemrinistic. They have a parameter to tune how stochastic they are.

Because they don't just parrot, they interpolate, which is why they have such varied abilities. You can't explain the range of behaviours they have with just parroting, and once you accept that, why shouldn't this qualify as some form of intelligence?

I don't see why it's those are the only two options, nor why they are even mutually exclusive.

> Spaced repetition is very effective, but it's really really clunky to use

Spaced repetition is just reviewing the same material periodically. It doesn't have to be a complicated system.

We already know how to learn and educate: spaced repetition (periodic review), and retrieval practice (frequent testing). This is how school used to be fore centuries; it's not sexy but it's effective.

> Humans can accurately retell what their consciousness was doing

Can they? How could we possibly know this is the case? People could simply post-hoc rationalize this to justify whatever decision they made.

Yes, computing in latent space is a big thing now.

https://ouro-llm.github.io/

nvfp4 is pretty good, and with 4 machines you get a roughly 4x speedup using tensor parallelism.

> I would expect the performance on a 768B parameter model spread across 4 to be significantly lower even though its a mixture of experts.

To fit that model you'd need multiple machines which lets you run tensor parallelism for additional speedups, roughly X speedup for X machines. Also you can't naively compare tps across models, and it would certainly run a nvfp4 quant or similar.

Politicians keep looking for excuses to break consumer encryption, and the youth social media harms are just the latest excuse that has received broad support. So disappointing.

Why? 4 x DGX sparks should be enough. That's way less than $80k.

> It depends a lot on things like geology and some people are a lot more sensitive.

People said this about high voltage electric lines and wind turbines. Blind tests proved they were imagining things.

Note that you're asking whether it's more impressive that someone managed to analyze and intentionally create an appealing, cross-cultural, and marketable product, rather than creating something appealing completely by accident. Of course the former is far more impressive than the latter, assuming it really was intention as laid out by the OP. It requires intelligence and understanding of the world.

Most bytes traversing the internet is video these days. Arguably, ffmpeg has processed most of that video.

I don't have to believe politicians, I see it with my own eyes whenever I deal with government employees.

Same way anyone else got their job, but that's beside the point. The point is that accountability in government is significantly lower than in private industry, and this is a big source of the problems.

> IMO wherever someone says "the government" we should mentally substitute "we all, collectively".

No, we should substitute "unaccountable bureaucrats". The people who enter and leave power from elections are not the source of the daily frustrations people have with government, it's the rest.

> LLMs are not that smart.

They are smart, but they are not aware of the environment they're in, or any implicit context that someone whose doing a job carries with them, that's why all of that context has to be explicitly laid out in a prompt. When the context is provided, they are quite smart.

> Why on earth would you go out of your way to do that? If someone wants to try it, why stop them? She just took it for granted that their job was to enshrine the existing state of things in a formal law.

This is exactly the Canadian experience: restrictions without thought.

> Newspapers have mechanisms like corrections and apologies that can be used to "right" a published falsehoods.

Ineffective mechanisms. So if we accept ineffective mechanisms as sufficient redress in those spheres, why not here too?