fizx

25 ms·

241.
by fizx 8y ago
IIRC, public companies have to disclose above $84MM, though perhaps not immediately.
242.
by fizx 8y ago
Science is the belief that the laws of nature are consistent, and therefore repeatable and testable. From there the practice of science is repeating and testing claims about nature, and communicating these tests through various social proce
243.
by fizx 8y ago
Just give everyone a month off. I see no problem with this :)
244.
by fizx 8y ago
I feel like that we're getting these huge gains on tasks that can be made faster via better architectures, regularization, normalization, data augmentation, etc, such that he's right. I just wonder if it will ever feel this way fo
245.
by fizx 8y ago
Dumbbell squats are safe, but I don't like doing as much weight as I do with bar squats, because I find 60+lb dumbbells very hard to manage. I also prefer using one dumbbell held with both hands (e.g. https://www.youtube.com
246.
by fizx 8y ago
Do body-weight squats to start, but do more of them. Then grab a couple 30lb dumbbells and hold them while you squat. If you can do 100 squats holding 30lb dumbbells, you are way ahead of the curve.
247.
by fizx 8y ago
Do you also not brush your teeth? Strength training can be as simple as doing one of squats, pushups, or pullups every day. Work towards 2-3 sets of 10-20 reps. Add weight (or reduce assistance) as necessary. You've just made a huge
248.
by fizx 8y ago
It's not surprising that an ensemble of embeddings beats single embeddings. I can't imagine it beats a single embedding fine-tuned to the task at hand, given sufficient training data. Where ensemble embeddings might shine is whe
249.
by fizx 8y ago
The article undersells how potent "spice" is--up to 50,000 times stronger gram for gram than pure THC. You're in the LSD world of micrograms having effects, and also the research chemicals world of not knowing what you'r
250.
by fizx 8y ago
If you aren't automatically rolling instances frequently, you will inevitably have some machines that were theoretically cattle but in practice are pets. (This is much easier today with k8s/mesos) More generally, when you claim i
251.
by fizx 8y ago
I built bonsai.io/websolr.com while working at Twitter in 2010. I left in 2012, keeping a promise to a co-founder that I'd leave if we hit $1M ARR. There were definitely a couple months of pager/ops overload where I hit 100h
252.
by fizx 8y ago
GitLab's "Auto DevOps" with GCP Kubernetes Engine is pretty good at the 1 to 100 story. I clicked several buttons and got a google-managed k8s cluster that will pretend to be Heroku for me. I've since outgrown the basic
253.
by fizx 8y ago
Sailing around the Mediterranean, then selling your boat two years later and sailing around the Caribbean, then doing it again somewhere else is easily within the reach of middle-class finances. Having a boat and crew large enough, comforta
254.
by fizx 8y ago
Video performance is a LOT harder than you think. At another company, I watched an five person team, half of whom had related PhDs, the other half over a dozen years of video engineering experience, make incremental improvements to the We
255.
by fizx 8y ago
I wonder what religious purpose future archeologists will think this served :)
256.
by fizx 8y ago
They sure hid figure 10! Unexpected section, bad header choice, etc.
257.
by fizx 8y ago
If you want concrete benchmarks, they compare to RocksDB and Redis around page 10 of their academic paper. ( https://www.microsoft.com/en-us/research/uploads/prod/2018/0... ) TL;DR: I find their choi
258.
by fizx 8y ago
Maybe its worth asking what properties a blockchain would need in order to provide a secure election, and ask whether those are possible. I'd love to see results for blockchain-related election technology with similar sorts of implica
259.
by fizx 8y ago
There are a lot of places in practical neural nets with attention where you want softmax(queryvector · memorymatrix), where memory can be quite large. If you have a decent ANN implementation, you can approximate by only calculating the dot
260.
by fizx 8y ago
One assumes its an extension of their previous work ( https://arxiv.org/pdf/1501.01062.pdf ), which was only valid for Euclidian and Hamming spaces? Edit: The author says its https://ilyaraz.org/static&#x
261.
by fizx 8y ago
Title is a bit clickbait, really.
262.
by fizx 8y ago
Actually, you should do better than 30% loss due to the non-linearity: 1 - 0.97**10 => 0.2625758731050719
263.
by fizx 8y ago
I wonder how far you'd get with a bot that macros perfectly but also A-moves 2-3 groups.
264.
by fizx 8y ago
In 2011 in Twitter, we wrote a prototype engine that would let you run tiny javascript programs in iframe "cards", which show up below the tweet where images do. The idea was that if you tweeted with #cardname, then twitter would
265.
by fizx 8y ago
If you have hobbies, look for a meetup around them. If you have pretty generic interests (wine, board games, and cheesecake), perhaps the best thing you can do is join a unitarian universalist church near you. It's all of the social b
266.
by fizx 8y ago
Last time I had a PEO employ me, I had at least healthcare and 401k matching.
267.
by fizx 8y ago
> some kind of paperwork-middleman will spring into place This is called a "PEO".
268.
by fizx 8y ago
Alternately, you could use a PEO! (i.e. completely outsourced HR services, such that your employees are actually employed by the HR's shell company) You get the management ease of contractors; they get the benefits of employment. It
269.
by fizx 8y ago
At least in the USA, its very hard to expatriate to avoid taxes. IIRC it requires a one-time payment of ~40% of your net-worth and you to have citizenship somewhere else. Other countries could follow suit, though the USA probably has the e
270.
by fizx 8y ago
This seems explainable. Humans typically learn art like GANs learn. We start as children making things and the adults, teachers, etc start grading and encouraging/discouraging us, providing the feedback we learn from. As we grow up, w

More ›