Razied

Wiki Contributions

Comments

As usual, being a bayesian makes everything extraordinarily clear. The mean-squared-error loss is just the negative logarithm of your data likelihood  under the assumption of gaussian-distributed data, so "minimizing the mean-squared-loss" is completely equivalent to a MLE with gaussian errors . Any other loss you might want to compute directly implies an assumption about the data distribution, and vice-versa. If you have reason to believe that your data might not be normally distributed around an x-dependent mean... then don't use a mean-squared loss

You're completely right, I don't know how I missed that, I must be more tired than I thought I was.

Answer by RaziedNov 16, 202210

Well, it will scale linearly until it hits the finite node-to-node bandwidth limit... just like all other supercomputers. If you have your model training on  different nodes, you still need to share all your weights with all other nodes at some point, which is fundamentally an  operation, it just appears linear when you're spending more time computing your weight updates than you are communicating with other nodes. I don't see this really being a qualitative jump, but it might well be one more point to add to the graph of increasing compute power dedicated to AI.

[This comment is no longer endorsed by its author]Reply

I don't think the bottleneck is iterative testing more importantly than in language/image models, all of them need iterative testing to gather the massive amounts of data required to cover all the edge cases you could care about. Your second statement about robustness seems correct to me, I don't think that self-driving is especially harder or easier than language modelling in some absolute way, it's just that the bar to deployment is much higher for cars because mistakes cost a lot more. If you wanted language models to be 99.999% robust to weird bugs, that would also be ridiculously hard. But if you want to hear the opinion of people right next to the problem, here's Andrej Karpathy, the (former) director of self-driving at Tesla, explaining why self-driving is hard.

Answer by RaziedNov 14, 20222420

It's mostly that the costs to self-driving failures are much higher than the costs to language models or image generator failures, when GPT-3 says nonsense, you can just shake your head and rerun it, you can't really do that in your Tesla. If anything I think that much more effort is going into self-driving, it's just that the statistical tails are really fat here, and there's a large number of unique situations that you have to get right before deploying them at scale. In short, it's mostly 2).

I completely disagree, stating that a person has status-seeking as a motivation makes some very clear predictions. For instance, we'd expect such a person to care very much about having their contributions recognized in front of their work peers, and we'd expect an event like a manager mentioning their idea in a meeting without giving them due credit to be very annoying to them. Conversely, if you observe someone who doesn't seem too bothered by not having their contributions recognized, you can state with high certainty that they are not seeking status. You'd also expect a status-seeking person to mention their accomplishments on occasion, if you see someone who never mentions any of their accomplishments to anyone, that is evidence that they are not seeking status. 

In general "status among humans" is very low-entropy, you don't have much status in most possible worlds, and you need to exert optimization pressure to steer the future towards the world where you do. I think the mistake you're making is equating "status = a human's value function", in which case the statement "humans maximize status" is equivalent to "humans behave according to their value function", which becomes tautological. But status does not capture the entirety of a human's value function, for instance, I care about being happy independently of having high status, I'd prefer both, but status is not enough.

Those who are conscientious and intelligent enough to be able to self-study university-level material and be consistent with it over a period of years don't need to attend this sort of university because they're already overwhelmingly likely to succeed in life the traditional way. And the rest (the vast majority of people), desperately need the structure of going to class to motivate themselves to do anything. And university is mostly not about learning anyway, it's an excuse to make friends, find future wives/husbands and generally have a good time and learn how to behave in adult social environments.  

Strong upvote. And well, your reply certainly builds up *my* trust of the rationalist community. Zvi should aim to maintain a healthy background level of disappointed readers to build up a robust peer-review population.

Ok, now for some nontrivial advice if you're actually optimizing for sexiness as a man:

  1. Develop your neck muscles by doing direct training (neck curls). The neck has a profound influence on your face aesthetics, see this video to see photoshopped examples and neck exercises.
  2. Develop your jaw muscles by chewing mastic gum for 30 minutes a day, again plenty of youtube videos that talk about this.
  3. If you can't grow a beard naturally, use minoxidil on your beard area, the gains are permanent even if you stop it after a few months.
  4. Read some redpill stuff like The Rational Male. You probably don't want to go all the way in that direction, but the attractiveness optimum lies in that direction with respect to your current position. 

Assuming an average daily consumption of 2000 kcal per person per day and a population of 330 million, that's around half a decade's worth of food from this stockpile alone.

And that 330 million number is likely to decrease significantly after a nuclear exchange sufficient to cause nuclear winter, so half a decade is likely a lower bound.

Load More