Akababa

9 ms·

31.
What Went Wrong at OpenAI? [audio](slate.com)
67 points by Akababa 7y ago|37 comments
32.
by Akababa 7y ago
Works fine for me developing on PyCharm on a new MBA. But on the other hand, I haven't used pre-10.15 macOS so I might be blind to some of the pain points of upgrading.
33.
by Akababa 7y ago
This particular article is about MFE students from one school, but hedge funds love PhDs too.
34.
by Akababa 7y ago
One way to get interviews at least is by doing well in math contests.
35.
by Akababa 7y ago
Right, so the issue with the income premium without economic growth argument is that the government would have collected those incomes anyway, just from different people (unless the student loans are structured as a % of future earnings, in
36.
by Akababa 7y ago
I feel like you're talking past a lot of the other commenters in this thread. I'll try to frame it another way: Generating value is synonymous with growing the economy and being more productive. You can't inflate incomes with
37.
by Akababa 7y ago
Nice, I learned about a couple new things. I take it these can all be implemented with the lock/semaphore primitives?
38.
How China Will Take over the World(forbes.com)
4 points by Akababa 7y ago|0 comments
39.
by Akababa 7y ago
That's true, but I don't think that makes it any less rational for them to diversify. 20% of your net worth is still a lot of risk to trade off just for a bit of investor confidence.
40.
by Akababa 7y ago
There are some good ideas in there, but I can see a few problems with this proof: 1. It doesn't show the result for all natural numbers, only some "high probability" fraction of them (similar to what Tao did). 2. The expected
41.
by Akababa 7y ago
Actually it's a fair nitpick because logarithmic density 0 does not imply natural density 0 - so the claim in the article is misleading in more than one way. For example, the set of numbers with a perfect square number of digits has lo
42.
by Akababa 7y ago
I wonder if some automatic governance model could be put into place to ensure a project never dies. Of course anything I can think of has the potential for abuse so maybe forking is the best solution after all.
43.
by Akababa 7y ago
I believe it has a lot to do with how visible the suffering is. If it's happening right in front of you and you can see it, hear it, etc. then it triggers a much stronger empathetic response than if it's separated by multiple laye
44.
by Akababa 7y ago
Who's matching the donations?
45.
by Akababa 7y ago
The world is moving towards cashless/mobile payments (albeit with the U.S. strangely behind), so why do you feel physical cryptocurrency is a step in the right direction? Do you expect Kong to face greater challenges for adoption (comp
46.
by Akababa 7y ago
Do virtual credit cards like Google Pay do something similar to this?
47.
by Akababa 7y ago
I was able to get it working without downloading files by mounting Google Drive, although once the shared folder is rate-limited you need to make a copy in your own Google drive: https://colab.research.google.com/github&#x2F
48.
by Akababa 7y ago
Google Drive is actually perfect for this as it lets you mount the drive (presumably as NAS) so you can read from it directly. I set up a notebook for it but unfortunately the shared folder got rate-limited. It still works if you make a cop
49.
One of Wall Street’s Most Lucrative Businesses Is at Risk(bloomberg.com)
1 points by Akababa 7y ago|0 comments
50.
by Akababa 7y ago
You can enter empty input to advance the generation.
51.
by Akababa 7y ago
If it's any help, I restarted the notebook and haven't been getting crashes for a while now.
52.
by Akababa 7y ago
I'm getting OOM errors on colab after the first 10-ish actions :( EDIT: Any chance you can support TPU on Colab?
53.
How the U.S. hydrogen bomb secrets disappeared(phys.org)
11 points by Akababa 7y ago|0 comments
54.
Qualcomm’s new Snapdragon 865 is 25% faster, comes with mandatory 5G(arstechnica.com)
6 points by Akababa 7y ago|1 comments
55.
by Akababa 7y ago
But we don't know how those 17 were chosen in the first place. There would be survivorship bias if models which match observations get more attention. With that being said, I doubt that survivorship bias is strong enough to explain the
56.
The rise and fall of the PlayStation supercomputers(theverge.com)
2 points by Akababa 7y ago|0 comments
57.
by Akababa 7y ago
I don't believe GP was making a point about ethics, but rather one of economics. In general it's not smart to advertise this to people if you want to keep your podcasts free for as long as possible.
58.
by Akababa 7y ago
You could say the same thing about adblockers in general. It's unfortunately pushed more content creators to put up paywalls and move to subscription-based models. The smart thing to do with apps like this is to keep it to yourself, bu
59.
by Akababa 7y ago
I don't know, this seems to be a really low-effort blog post. The given example is obviously contrived from the unreasonable improper (-\infty,\infty) prior and the low \sigma^2=1 likelihood. If it was really "pure noise" the
60.
by Akababa 7y ago
What do you mean by coordinate transformation? MLE is invariant under parameter transformations because it's just the argmax of the likelihood.

More ›