The average household also includes empty nesters and people who have moved out yet haven't started a family yet. The average number of children per family in the 60s was 2.3, which means quite a few families had 3 or 4 children.
But what is the purpose of "scaling?" Broadly speaking, it means that there is a marked and obtainable improvement on an objective function in response to additional data. But for certain criteria, like the possession of knowledge, it appears attention-based LLMs aren't capable of that, regardless the amount of data. So they don't "scale" if you use that as an objective function.
No OP, but no amount of data will ever enable an LLM to know when it doesn’t know something, for instance. That’s a fundamental limitation with the current architecture.
Imagine a world where a single OSS maintainer can do the work of 100 of today’s engineers thanks to AI. In the world you describe, it seems likely that contributors would decrease as individual productivity increases.
This was a few years ago, but blob storage on GCP had a global outage due to an outage in a single zone. That, among numerous other issues with GCP, lost my confidence entirely. Maybe it’s better now.
> Why is that? The odds of being right are 50-50, so why not switch?
Couldn't you ask the opposite question: if the odds are 50-50, why switch? Keeping your initial choice is more passive than switching. If the odds of success are the same, shouldn't the choice with the least energy expenditure be preferred?
Fundamentally, it’s hard to enforce certificate pinning if the user can modify the binary. Even if sandbox mode used certificate pinning, there would likely be some other way of removing the pinned cert checks.
In many cases, you don’t really need to check pointers unless you specifically want to handle the nil case. Go’s panic / recover feature lets you “catch” nil pointer access elsewhere. For instance, if you have an HTTP handler, it’s way easy to install a middleware that recovers from panics than it is to check for nil everywhere.
Another way of phrasing this is 23% want a job earning more than 25K per year and are incapable of finding one. I can’t imagine living on 25K per year. While I would technically be “employed” I wouldn’t be happy about it. What is the purpose of measuring employment? I think satisfaction with one’s living circumstances is certainly part of the reason. From that perspective, the measure proposed in TFA makes some sense.
I don’t think you understand what you’re saying. “A linear combination of non-linear features” is not what people mean when they talk about linear regression. And even if they were, a neural network will do a much better job of generating the non-linear features than you will be hand. So again, it’s logically obvious that a neural network will be capable of performing better than simple linear regression.
It’s logically obvious. You can recreate any linear regression using a neural network. So a neural network approach will always be at least as good as linear regression. But a neural network can model non-linear relationships as well. Now ask yourself, what is the likelihood of there being a non-linear relationship in the data? As with most real-world data, the likelihood of non-linear relationships is extremely high. That’s why we can be quite certain a neural network approach will able to outperform linear regression.
> Are these deep learning detectors better than running a simple regression after using something like a wavelet transform to detect these specific features?
It’s pretty insane to think that a decent neural net trained on sufficient data wouldn’t be able to outperform basic regression.
I don't really think it's appropriate to draw conclusions from studies that control for caloric intake. Most people don't treat food as fungible calories. Often, drinking juice is additive in calories.