ma2rten

73 ms·

811.
by ma2rten 8y ago
It took me a moment to understand what kernel in the title is referring to. They are not referring to the machine learning term, but a CUDA kernel, a GPU program.
812.
by ma2rten 8y ago
I don't think the point is that Bill Gates should be ashamed. It's a reminder how fast technology changes and that the assumptions we make today could be invalidated just as quickly.
813.
by ma2rten 8y ago
MySQL used to be more popular. It was the M in the LAMP stack, which virtually everyone used to use. It also used to be more scalable, because PG's replication story used to be poor. This is why Google and (I assume) Amazon used MySQL
814.
by ma2rten 8y ago
Most of the work was done during two sprints, for a total 2-3 person-months of work. I think what they are saying is just: We got a lot of work done on this during those two sprints. It's not a statement about the work that is remain
815.
by ma2rten 8y ago
Not necessarily. In the case of word vectors we are using unsupervised learning to identify patterns in a large corpus of data to improve the learning. This is a completely different issue than your credit score example, which is supervised
816.
by ma2rten 8y ago
I also disagree with the conclusion, but for a different reason. I think it's unlikely that the word embeddings were just lower quality. That should result in noise, not bias. I that there is a real statistical pattern in the training
817.
by ma2rten 8y ago
I was just trying to explain in this comment why I think the human moderation solution is solving the wrong problem.
818.
by ma2rten 8y ago
I think that the bias problem they are highlighting is very important. That said, I'm wondering if they really didn't try (like the title suggests) or if they choose this approach on purpose because it highlights the problem. To e
819.
by ma2rten 8y ago
This approach only works if you use OP's assumption that a text's sentiment is the average of it's word's sentiment. That assumption is obviously flawed (e.g. "The movie was not boring at all" would have negati
820.
by ma2rten 8y ago
People used to think that Chrome will never take over IE on Windows. Most people will just keep using the default browser. I think that was valid. I don't really understand how Google managed to get people to switch. Is it just ads on
821.
by ma2rten 8y ago
This article has more background: https://mashable.com/article/google-location-history-trackin... Essentially there are multiple options for location history and the labels are confusing.
822.
by ma2rten 8y ago
I don't know what the situation was, but maybe you should have let them know that you were going to write a python script?
823.
by ma2rten 8y ago
You and OP might not be typical users.
824.
by ma2rten 8y ago
Are you sure that that is actually a big problem for them? I imagine it would be quite hard to convincingly fake user data data. I could also imagine ways around that. For instance their proprietary SDK could generate the data and cryptogra
825.
by ma2rten 8y ago
now that the commonly used hostnames are published. The problem I see with that is that apps could easily start proxying the requests though their own servers.
826.
by ma2rten 8y ago
Also when you click on the link it says "handbook for new employees".
827.
by ma2rten 8y ago
Maybe it could be like YouTube where the creators get part of the ad revenue?
828.
by ma2rten 8y ago
I read the full article in question. It was indeed very interesting. The meat of the article were five specific issues that Waymo cars still have. I hope theinformation forgives me for hot linking it: https://dmxvlyap9srmn.cloudf
829.
by ma2rten 8y ago
I feel like Santa Clara county has a bunch of these dangerous merging lanes.
830.
by ma2rten 8y ago
You could use clustering, dimensionality reduce or feature selection. However, the way I have seen the hashing trick being used is not to compress the feature space. For most problems it would be a bad idea to just lump your most discrimina
831.
by ma2rten 8y ago
Genuine question: Do other high-skilled professions have unions (like marketeers, scientists, lawyers, doctors, electrical engineers, ...)?
832.
by ma2rten 8y ago
The hashing trick (a la sklearn's dictvectorizer) can make a huge difference. I'm not a huge fan of that. Hash collisions can lead to unexpected behaviors in production and make feature attribution for debugging harder. It'
833.
by ma2rten 8y ago
Tech companies used the same strategy in the 1970s, offering employees high salaries and sweet perks to make collective action less appetizing. That doesn't make sense to me... I think it's more likely that tech workers get high
834.
by ma2rten 8y ago
I'm surprised that this comment was downvoted so much. I actually work at Google. I don't know if anyone is still going to read this, but I thought I would comment just in case. I said "for the most part". There are mult
835.
by ma2rten 8y ago
That's not true either, but that's a common misconception. Google makes the bulk of it's ad revenue from search ads. Two people issuing the same search query will see the same ads (for the most part). Hence, Google could stil
836.
by ma2rten 8y ago
Google docs has that feature: https://support.google.com/docs/answer/6367684
837.
by ma2rten 8y ago
I don't read a lot because I have ADHD which makes it hard to focus. Do you also have that problem?
838.
by ma2rten 8y ago
I'm on Hacker News since eight years. My life would not be the same without it. A lot of what I know I know through Hacker News. HN has shaped my world view. Over the years I learned about starting a business, databases, programming la
839.
by ma2rten 8y ago
You misunderstood what I mean. I will edit my reply for clarity.
840.
by ma2rten 8y ago
I saw that and I don't see an issue. Money (unfortunately) influences politics in the US and the best way to influence it is to give across the political spectrum.

More ›