vadarvariu

CelebA is a relatively standard dataset that is used to train DL-based CV models, it's around 200k of celebrity faces. While the article title is no doubt clickbait, it wouldn't surprise me if Gosling is indeed lurking in those model weights somewhere.

Now, consider that this is the cost of the final model reported in the paper. This doesn't account for all the iterations of trying out e.g. different model architectures, hyperparameter sweeps, etc. The true cost of the experimentation is likely at least an order of magnitude higher.

My personal monthly usage is in the 100G region; this particular failed update has cost me 20G of burned bandwidth. Where I live around 2% of the country doesn't have access to decent broadband. Since these updates are released every 2 months or so in a very back-of-the-envelope fashion failed updates amount to 0.2% of the country's total bandwidth usage. I might be an order of magnitude off but that is still an incredible waste.

Maybe they did run the numbers, but the amount of engineering effort required to fix it is absolutely minimal (couple of lines...). A significantly worse outcome for society as a whole at the cost of a try-catch block.

I've had the opposite problem: started getting awful freezes on a MBP 2018 running Mojave on the day this latest Catalina version was released. No immediate explanation: cpu/ram/disk load perfectly normal, machine just incredibly slow (think 10 minutes to start up). Exact same behaviour in Safe Mode and nothing obvious after some common diagnoses.

Bit the bullet and finally updated to Catalina, all problems gone. Not sure what to make of this but it definitely resonates with stories of Apple practically forcing updates.

Totally agree. Updated to Catalina recently and on top of the huge patch size the update client doesn't seem to have any retry mechanism built in. Connection drops 9.9G into a 10G download? Too bad, start over. Just a simple wget -c would do. Incredible.