I disagree. The scientifically interesting thing is that we can have such exploding complexity, while at the same time every step of the algorithm still actively works towards a solution.
The article compares the return on the cost/investment of hiring a rock star vs NOT hiring a toxic. It acts as guidance on what part of one's hiring pipeline is most worth investment.
I took the Overland a few months ago and discussed it with the train crew...after having to wait for a freight train.
When I was a kid we used to take the sleeper trains on that route but these days it's pretty much entirely freight. Most of the country stations have been shut.
You are correct, I misattributed the top comment as part of the article, was an honest mistake.
"Old growth forests tend to exist in a carbon steady state; younger, growing forests tend to be net carbon uptakers. This situation is due to natural disturbance regimes in forests, like fire and pine beetle, having historically signifiant impacts in large part because of climate change."
More critically, they emit huge amounts of sulphuric compounds, which are what really sick around and tend to cooling. Unfortunately when those same compounds don’t reach high altitudes, you get a lot of acid rain. All told however, their net effect is strongly cooling over the first few years, and then milde cooling over longer time frames.
Why cooling though? It makes sense that it cools the ground, but the sunlight is still captured by the particles (as opposed to reflected by snow/water, on average), so the total Earth+atmosphere should be warmer?
Fantastic, thanks for the explanation. Can you please point me to a source? Need to send it to my dad who taught me everything I know but has now turned into a climate change denialist who says it's all a globalist / marxist / leftist plot to make money off solar panels.
Thank you so much! I will send them all three. Just because you offered and you seem so knowledgeable, he has also claimed the CFC damage to the ozone layer was a hoax created to promote alternative refrigerants... If you have anything on that I could use a link as well!
Some readers seem to be getting thrown by the casual tone of this article. For those struggling to not rage my two takeaways were:
* Automated checking of rusts std lib could improve rusts security
* Don't use unsafe if you don't need it
* Releasing a fix for a security vulnerability should be complemented with a cve if you want people (such as anyone using Debian) to not still be vulnerable two years later
Note: despite the initial slant, the author is very pro rust.
I wouldn't even say the initial slant is anti-Rust. To me it reads as "Rust wants what I want, which I appreciate, and that is exactly why I won't go easy on it". Yes, the title is click-baity, but author also immediately acknowledges that "unsafe" is a necessary compromise:
> But dealing with those things is necessary to run code on modern hardware, so something has to deal with it.
The main point is that "unsafe" is the weakest link in the language, since it bypasses the safety guarantees of Rust. Therefore it needs solutions outside of the language (in the form of how bugs are handled, and programmer culture) to minimize the risks it brings.
You have missed the point. Nothing in the study mentions a higher rate of reported cannabis use than in the general population. The record scratch is due to implying generalisations based only on a non-typical population of infertile couples without any control.
Project the data, not the model. You need to do it anyway and has the side benefit of making overfitting more difficult. Protecting the model is impossible anyway since most advancements terms to be trying lots of publicly known techniques and discovering that a particular combination works best for your data. Once one knows which techniques those are these nothing stopping a competent engineer from reimplementing them at another company.
Indeed. For commercially useful applications, collecting the data, labeling it, etc, costs orders of magnitude more than a team of PhDs building the models.
> For commercially useful applications, collecting the data, labeling it, etc, costs orders of magnitude more than a team of PhDs building the models.
I don't think it's typical. For example, JFT has 350e6 images, and it probably cost ~$35M to hand-label, but Google has paid people far in excess of that to work on image classification.
>Of course, the elephant in the room is where can we obtain a dataset that is 300x larger than ImageNet? At Google, we have been continuously working on building such datasets automatically to improve computer vision algorithms. Specifically, we have built an internal dataset of 300M images that are labeled with 18291 categories, which we call JFT-300M. The images are labeled using an algorithm that uses complex mixture of raw web signals, connections between web-pages and user feedback. This results in over one billion labels for the 300M images (a single image can have multiple labels). Of the billion image labels, approximately 375M are selected via an algorithm that aims to maximize label precision of selected images. However, there is still considerable noise in the labels: approximately 20% of the labels for selected images are noisy. Since there is no exhaustive annotation, we have no way to estimate the recall of the labels.
That doesn't sound like recaptcha: it's more likely that they label the pictures N (or n%) people click after searching for "Golden Retriever" in image search (as the "raw web signal")
