Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
whimsicalism
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
2435 ms
·
8611.
▲
by
whimsicalism
6y ago
Sure - also the metrics measured in industry are not the same as the metrics measured in school. What you're saying doesn't seem at odds with what I'm saying. I would, however, be interested in seeing a study where they give
8612.
▲
by
whimsicalism
6y ago
From my perspective, the thing that has hurt the credibility around ADHD diagnoses is the pretty blatant use by affluent parents to get diagnoses so they can get test-time extensions for their kids. ADHD diagnosis is much more strongly corr
8613.
▲
by
whimsicalism
6y ago
Imagine if we put even one iota of the effort we put into fighting petty theft & drug crime into fighting employer crime. About equal amounts of money are stolen through wage theft as through petty property crime. Yet you don't see
8614.
▲
by
whimsicalism
6y ago
Democracies are far too short-sighted for this to be an actual reason. Also, this is just a paper published in a journal.
8615.
▲
by
whimsicalism
6y ago
There's just no way that we as a society when deciding between "some shiny new satellite thing produced by an American corporation" versus "the quality of the VLAs operations" will pick the latter. It pains me to sa
8616.
▲
by
whimsicalism
6y ago
I mean even labor laws on the book are hardly enforced in the United States.
8617.
▲
by
whimsicalism
6y ago
Yes, but China only became rich enough that lots of wealthy young Chinese could be buying iPhones recently. I do not think it would be unreasonable to make the same bet on India being a decade or two behind.
8618.
▲
by
whimsicalism
6y ago
> I do not want any App Store to be taking political stances. I don't think there is a clean separation between political and non-political when you're choosing whether to accept or deny apps to an app store.
8619.
▲
by
whimsicalism
6y ago
> may just end up spiraling into a hot bureaucratic mess Many publicly traded corporations are not that much different. I am confident that an app-store is not that hard of a task relative to the revenue Google et al are extracting from
8620.
▲
by
whimsicalism
6y ago
What do you mean? I'm familiar with Kaldi - although haven't tried running it on an Rpi (nor with these particular models). That's materially different from w2l though.
8621.
▲
by
whimsicalism
6y ago
> Comparing treament of Uighers to caste situation in India is unfair. Absolutely, not my intention by my comment - more just recounting a similar phenomenon in a different discussion. > India is fighting it on policy and social level
8622.
▲
by
whimsicalism
6y ago
What a surprisingly condescending response, I thought you were genuinely curious. I never said anything about lower accuracy - we are running full size transformer models for translation. And wav2letter++ inference models are SOTA on the Li
8623.
▲
by
whimsicalism
6y ago
That would require distributing the trained AI model weights with your compressed text, entirely defeating the space gains for all but extremely large passages of text.
8624.
▲
by
whimsicalism
6y ago
Do you know of any performance/cost benchmarks that you could point me to showing this?
8625.
▲
by
whimsicalism
6y ago
Sure, we are not using any advanced parallelism beyond just batching requests for translation and spinning up more servers if we have more requests. No distillation. For translation, there's really not that much to say - we run the tra
8626.
▲
by
whimsicalism
6y ago
So because a few black people make money off of sports we don't have a racially biased system? There are also rich "low caste" people in India - it doesn't mean there isn't caste discrimination.
8627.
▲
by
whimsicalism
6y ago
> fun times. doesn't sound like it
8628.
▲
by
whimsicalism
6y ago
To be very clear, I'm not claiming the practice is just. It isn't. My only claim is that this doesn't arise just because of bias of the interviewers.
8629.
▲
by
whimsicalism
6y ago
I work for a for-profit public SaaS corporation, we are not optimizing for energy efficiency, we are optimizing for cloud compute costs.
8630.
▲
by
whimsicalism
6y ago
> For some task inference CPU can't be real time. Example for some of those task are speech recognitions and friends, machine translation etc Funnily enough, the two services that my team is running on CPU are speech recognition and
8631.
▲
by
whimsicalism
6y ago
It's odd - often when I talk with international people, they seem to be under the impression that affirmative action doesn't exist outside of their own country. I was recently talking with someone in China who claimed that the Uig
8632.
▲
by
whimsicalism
6y ago
Are people largely running inference on GPU/TPU? At my job, we run inference with pretty large transformers on CPU just because of how much cheaper it is.
8633.
▲
by
whimsicalism
6y ago
I mean there are two effects: 1. Ongoing discrimination 2. Different starting points + law of large numbers. If you start one racial group off in a country with property and let them acquire wealth for a few hundred years, then free another
8634.
▲
by
whimsicalism
6y ago
Absolutely, but I don't think it diminishes the necessity of the HR report.
8635.
▲
by
whimsicalism
6y ago
> Ivy League schools in industries like finance This also has to do with the fact that, in consulting and finance, it is the perception of intelligence by the customer that is much more important than the actual intelligence. Managers wo