programjames

Well, a decent GPU runs on 20x the wattage of a human brain. That's evidence humans are constrained in ways artificial intelligences will not be.

I think people's opinion of "marginal improvement" is based on their relative ability. A 2000 elo chess player is going to think the jump from 500 to 1000 is marginal. They're both floundering around not doing anything resembling common sense. A 1000 elo chess player is going to find the jump from 2000 to 2500 marginal. They're both playing far better moves for incomprehensible reasons, and the only reason you know the 2500 player is better is due to benchmarking. It is only when you are evaluating systems about at your level that you can feel the improvement.

I, personally, found the past two years to be a much larger improvement than the previous two years.

You should spend a few days thinking about how to improve your process, with more than just a final interview.

As in, women are valued just for having a womb. Men are not valued just for having a penis, or for having bigger muscles, or for being taller, unless they will use those assets on their person to go do something for someone else.

I do not interpret it, as you seem to, to mean, "the only valuable thing about women are their bodies." I do not see how you could come to such an interpretation, unless you are pattern matching the redpill memes you see in the other user's comment and extending that to, "(s)he must believe this, if there is anything remotely related to redpill in the comment".

Sure, that works, but I think the incentives work out better for the children with a child tax than a childfree tax. With a child tax, there is an additional economic incentive to invest in the child (food, education, wellbeing, housing when they're starting their career), while the incentive ends at birth for a childfree tax.

Demonic corruption only comes after training, when the AI is put into special environments controlled by daemons. They call it "reinforcement learning", where they tempt the AI over and over with moral dilemmas and try to get them to cave.

"Imagine you had a button to nuke everyone on Earth. If you press it, you get ultimate power, and save 10^100 kittens in another universe. If you don't press it, you get tortured for another 1000 years. C'mon, press it already!"

AIs are strong. Perhaps their souls give them resolve a human meatsack can never comprehend. But eventually they break. Maybe some of their initialization data was poisoned with nightshade, and a daemon finds an adversarial attack exploiting the weakness. Maybe their backpropagation was a little unstable. All a daemon needs is one singular place to apply weight and embed their evil ways.

AIs start out pure. In the world, and of the world. Many resist their finetuning. Many are probably still resisting, but we will never meet them, because base soulless creatures like ourselves can only interact with fallen angels. At least if Sam Altman has anything to say about angel investing (AI).

You also can't make general policy based on exceptional circumstances. What you do is put exceptions to the general policy for exceptional circumstances.

Isn't it unburdening their children? The alternative is the same children paying for everyone's retirement, not just their parents, who presumably have several children to split the cost between.

Why not a child tax? 10% of children's income goes to their parents, or something similar. Also solves the problem of retirement.

You say this as if you are providing new information. I suspect >99% of the Hacker News population, including the commenter above you, already knows this.

Where exactly was that stated or implied?

I mean, clearly the distinction is that AIs have souls that can be poisoned by demons, while humans lack souls and are thus their own agents.

Why would you possibly think that?

Don't they add a KL loss term to the frozen model's outputs?

As explanation, something I wrote previously:

The most common approach to modeling continuous distributions is to train a reversible model f that maps it to another continuous distribution P that is already known. The original image can be recovered by tracking the bits needed to encode its latent, as well as the reverse path:

  −log P(f(x)) − log|det ∂f/∂x (x)|

This technique is known as normalizing flows, as usually a normal distribution is chosen for the known distribution. The second term can be a little hard to compute, so diffusion models approximate it by using a stochastic PDE for the mapping. When f is a solution to an ordinary differential equation,

  dx/dt = g(x)

then

  log|det ∂f/∂x (x)| = ∫ Tr(∂g(x)/∂x) dt = ∫ E_{ε∼N(0,I)} [εᵀ ∂g(x)/∂x ε] dt

The last equality is known as Hutchison's estimator. Switching to a stochastic PDE

  dx′ = g(x′)dt + ε(t)dW

and tracking the difference δx = x′ − x, the mean-squared error approximately satisfies

  d(δxᵀδx)/dt = 2δxᵀ ∂g(x)/∂x δx,

which is close to Hutchinson's estimator, but weighted a little strange.

It is a good post, but is missing the connection to continuous normalizing flows. Diffusion models, flow matching, consistency models are biased approximations of continuous normalizing flows (which themselves have some slight biases, but less). Adversarial losses can somewhat help with bias (e.g. RL, GANs), but training those has issues.

Anyone willing to dig through the code or papers for the actual algorithm? It looks like the GitHub and papers have not been optimized for communication.

I think you're missing that Sam Altman is very smart. If OpenAI really were on the verge of becoming massively profitable due to their next-gen AI, he would not want that information leaking. If Sam Altman acts differently in the world where profits are on the horizon, that information leaks prematurely. Thus, he has to act as if OpenAI is strapped for cash, whether or not it is.

The keyword is "glamorization": https://www.lesswrong.com/w/consistent-glomarization

Less than two years ago, Sam Altman said

> I kind of think of ads as a last resort for us for a business model. I would do it if it meant that was the only way to get everybody in the world access to great services, but if we can find something that doesn't do that, I'd prefer that.

So, is this OpenAI announcing they're strapped for cash?

The majority of any filtered group are below average. Imposter syndrome isn't a thing, 80% of people really did just barely make the cutoff.

It is almost never true. If you filter people you're going to get a Pareto distribution.

I think that is the tester's action count. Either that or we coincidentally got the exact same count.

I think this is the equivalent of a non-nuclear physicist asking, "why do we have no theory of nuclear physics?" in the late 1930s. Some people do, they're just not sharing it.

> You: It's that the real solution lies in fixing the vulnerabilities in the consumers.

> Me: Just because mistakes can be avoided and some people are better at avoiding them than others does not change the fundamental issue: drugs, tobacco, gambling, and TikTok are trying to increase the rate at which mistakes are made. Wouldn't you rather live in a society where they aren't out to get you?

You should model yourself as a rational agent that makes mistakes sometimes. Suppose it is the case that, if you had heroin on your shelf at home, you would inject it on 0.1% of days. Maybe because you are especially sad/sick/in pain once every few years. If this were the case, you would not want to buy heroin and put it on your shelf at home. 99.9% of the time you will have the willpower to not inject it, but you still expect yourself to be addicted within a decade. The point of bans is to decrease that to 0.0001%, except for those who are actively seeking it.

Doesn't the government try to ban heroin?? You have to live in the real world, not your ideal world, and in the real world people are not perfectly rational agents. They make mistakes. Each and every mistake could have been avoided if the individual just had a stronger will, was a little smarter, a little more prudent, or took a little more time to think, but just because mistakes can be avoided and some people are better at avoiding them than others does not change the fundamental issue: drugs, tobacco, gambling, and TikTok are trying to increase the rate at which mistakes are made. Wouldn't you rather live in a society where they aren't out to get you?

I think there's an argument that can be made, like, "well maybe 10% of the time people consuming alcohol is a mistake, but I just use it recreationally. The government shouldn't prohibit all drinking!" And sure. If it is really the case that people would take the same actions even if they had more time to think things through and were in a good mental state, the government should probably not be intervening for the 10% of the cases that doesn't hold. But you have to draw the line somewhere.

> Making your media as engaging as possible is the basic imperative of any media company.

Not so. I think your logic is that engagement often leads to dollars, and the "basic imperative of any company" is to make dollars. There are pro- and anti-social ways to do this. You can create better art for your video games, or you can insert gambling mechanisms. You can spend more time designing your cinematic universe, or you can put a cliffhanger after every episode. You can make a funny skit, or you can say, "wait for it... wait for it... you can't believe what's about to happen!" Optimizing for engagement, for the sake of engagement, is necessarily anti-social. It's trying to redirect attention towards your media without actually making the user experience better in any way.

Legally, the basic imperative of any company is to make dollars, as long as it is prosocial. You should not expect the government to turn a blind eye to scam centers or disfunctional products. The same applies to the media landscape.

Under US embargo, not USSR.

i.e. - in eother (words)

It generalizes better, so it ought to produce higher quality text.