Mehdi2277

14 ms·

91.
by Mehdi2277 5y ago
I used to work at tiktok on recommendation. While we definitely do better exploration than some places there’s still a very clear top heavy bias. Part of that is just content quality. A lot of content is worse than others for any form of me
92.
by Mehdi2277 5y ago
Even if we had a demographically ‘fair’ recommender, a normal recommender is super top heavy. Telling 1000 companies person x is great is not very useful to them or the rest of candidates. I would expect this system to need some heavy limit
93.
by Mehdi2277 5y ago
The inference system and the type system are separate. Pyright uses a very different inference algorithm than mypy for better incremental support. It remains pep 484 compliant like mypy. Pyre's taint analysis is separate from pep 484 a
94.
by Mehdi2277 5y ago
I've had 3 jobs since college. First one I unintentionally negotiated the internship although did not negotiate the later full time conversion. 2nd/3rd jobs both were negotiated. 2nd job was around 20% increase while 3rd job was 1
95.
by Mehdi2277 5y ago
This is commonish in translation community. There are sites like wuxia world and woopread that have a lot of translations of east asian web novels that you can pay to see chapters earlier. Some do it as pay per chapter others as a subscript
96.
by Mehdi2277 5y ago
There are smaller similar sites to wattpad that are easier to get viewers for new novels. I like tapas, but probably several more worth exploring (unsure if woopread is only translations or supports self publishing). I'd likely submit
97.
by Mehdi2277 5y ago
Not that much as they all follow pep 484 + a few other peps. The number of type features unique to each one are generally bugs/recent peps they haven't implemented. The biggest one is mypy plugin ecosystem although the hope there
98.
by Mehdi2277 5y ago
In practice they are quite similar in errors reported. There are differences that pop up but most of the time either it is bug to report or one has implemented a certain pep faster than another. A few rare things are extra type checking sup
99.
by Mehdi2277 5y ago
FLOC in the spec allows user to disable it. So even with chrome you can always go to the privacy settings and disable floc directly for yourself.
100.
by Mehdi2277 5y ago
Google does and the cohorts are opaque and data generated based on history. They're not actually using this specific approach but roughly what they're doing is like taking your browser history, representing it as some vector of nu
101.
by Mehdi2277 5y ago
There are many work conversations I have that rely heavily on threads and make the channel useless to scroll if sent directly. I think most of the discussions I read are in threads as often one engineer will encounter a bug, ask for help, a
102.
by Mehdi2277 6y ago
The voxel bounded neural radiance field is important as neural radiance field was some prior research paper this builds off. But the very high level is just voxel data to image generation using some form of neural nets. I didn’t look at the
103.
by Mehdi2277 6y ago
Data classes do not do runtime type checking. They are a way for static type checkers to do validation. So there’s no issue for data classes in 3.10.
104.
by Mehdi2277 6y ago
It’s in multiple places. The main discussion place for python core development is in the mailing lists and there are topics for this issue posted there today too.
105.
by Mehdi2277 6y ago
For political content I was thinking more of qanon content. Also you can bury it in the recommendation sense but you can’t bury it in the search engine sense without removing it. The mere existence of qanon like content on the platform even
106.
by Mehdi2277 6y ago
The type system is specified but the main pep people point to is the base type system. There have been multiple extension peps adding new type features and different type checkers have different progress implementing them. Also different ty
107.
by Mehdi2277 6y ago
You determine which is right and report a bug in the other. After that you add a type ignore statement to keep the wrong one happy until the issue is resolved. My current workplace I have two type checkers enabled (mypy/pyright) and th
108.
by Mehdi2277 6y ago
That’s because pyright was written in typescript so it makes sense to use npm to install it. Pyright has weak ide support beyond visual studio stuff. I’m pretty fond of pyright due to how responsive the developer is on GitHub. I’ve made mul
109.
by Mehdi2277 6y ago
My experience having worked at content platforms on ml (so fb, tiktok, snap) the largest motivation for content removal is content that general audience would be uncomfortable with. Pornographic content is the easiest example. Should a vide
110.
by Mehdi2277 6y ago
Quants still pay a silly amount that being a quant at a top firm (citadel, jane street, HRT, etc) still out pays FAANG most of the time. A candidate that interviews well and negotiates some can go for 300-400k new grad bachelors to those fi
111.
by Mehdi2277 6y ago
I don't think mathematica supports AD. It supports symbolic differentiation which is similar, but different. Symbolic differentiation given a function determines the derivative for the whole function. AD focuses on running a function a
112.
by Mehdi2277 6y ago
Porn type content also tends to have trouble with many services like payment processors. A simple example is paypal doesn't support most porn related content. I also recall a video game company that makes weirder porn (stuff like hypno
113.
by Mehdi2277 6y ago
The simple answer is the age penetration in older ages has grown. You can use these numbers to get a rough estimate, https://www.statista.com/statistics/1095186/tiktok-us-users-... Disclaimer: I work at US tiktok.
114.
by Mehdi2277 6y ago
They've already done that. Currently there are a couple major data centers being in US/Singapore. US data is not permitted to go to china. The data currently is stored in a mix of US/singapore so can't claim solely US. T
115.
by Mehdi2277 6y ago
The simplest spam filtering algorithm would be a naive bayes filter. It's essentially keep a count of words that appear in all posts, words that appear in spam posts, and words in non spam posts. Those counts + bayes rule will let you
116.
by Mehdi2277 6y ago
I'm a similar age as you, but personally wouldn't count 9/11 (although maybe the year difference is why). For me 9/11 happened when I was in pre-k. I don't remember learning about it back then and had no clue what 9
117.
by Mehdi2277 6y ago
UCL is another good choice for this area. I did some research in this area back in undergrad. There was a workshop called NAMPI (neural abstract machine program intelligence) on this subfield so I'd recommend going through the accepted
118.
by Mehdi2277 6y ago
You're aware around 90 percent of the bay area considers to homelessness to be a serious problem. Citation: https://www.mercurynews.com/2020/03/08/when-it-comes-to-the-... "89% of residents say home
119.
by Mehdi2277 6y ago
Google outpublishes most (maybe all) universities if we restrict to AI research. Not particularly surprising given they have several hundred AI researchers and likely over a thousand (deepmind alone is a couple hundred). They also do intere
120.
by Mehdi2277 6y ago
My experience applying for a phd and talking to professors is it has gotten a lot more competitive. When I applied a year ago to 8 schools my recommenders thought I'd picked enough schools and had a strong enough background for top sch

More ›