emcq

11 ms·

61.
by emcq 10y ago
That's a cool experiment, but from a quick read they are creating a compiler which is a bit different than web development. Looking at the related work all seem to contradict this finding and suggest preference to statically typed lang
62.
by emcq 10y ago
Even if this only allowed device based training and not privacy advantages it's exciting as a way of compression. Rather than sucking up device upload bandwidth you keep the data local and send the tiny model weight delta!
63.
by emcq 10y ago
This is awesome, thanks! My apologies I must have missed it somewhere.
64.
by emcq 10y ago
If the main contribution here is the quality of the model and its interesting and powerful representation of text, I hope OpenAI does something distruptively different and releases the weights and trained model. The accidental sentiment neu
65.
by emcq 10y ago
It's very difficult to understand what the contributions are here. From what I've read so far this feels more of a proposal for future research or a press release than advancing the state of the art. * Using large models trained o
66.
by emcq 10y ago
What you describe is exactly what practitioners in the field have been doing for years. I think that's why the parent is a bit puzzled at the publication, as it's difficult to understand what's novel.
67.
by emcq 10y ago
The technique of training a model on a lot of data for a long time and then leveraging its sophisticated representation only learning the last layer(s) on small datasets to create accurate models is common practice.
68.
by emcq 10y ago
Why bother with rounding errors in the budget when every large spending has some biased influence on the population? Of course it's much harder to make real effective changes to the budget than bicker.
69.
by emcq 10y ago
There has been a microcode cache after the decoder since at least the Pentium 4. In the Pentium 4 it's called the trace cache. Check out "The Microarchitecture of the Pentium 4" if you're curious about the architecture.
70.
by emcq 10y ago
Up until the last sentence of the grandparents post, the point was "I think people are unable to challenge stars in their field." The fact that an alternative perspective that proves simple, effective, and popular in alternative c
71.
by emcq 10y ago
This is so true. We can see it very recently too. As a specific example look at evolutionary optimization. It's existed for decades, has a bit of a non mainstream cult following, but now the leaders of the field are finding they can be
72.
by emcq 10y ago
There are 540 billionaires in the US. The top 400 combined are worth 2.4 trillion. That's a remarkable portion of the economy.
73.
by emcq 10y ago
I think the parent is trying to make the case that startups by nature can't provide competitive healthcare, which hurts the economy when small business can't attract top people. It's much different than a 401k, free meals, fo
74.
by emcq 10y ago
From my own experience, it was naivety and manipulation. I trusted the founder when he would make big promises for the future. He didn't bring up equity until the last moment once we already were ready to quit and join. In hindsight th
75.
by emcq 10y ago
Having recently updated from 10.8 to 10.12 I can tell you it wasn't happening before, but definitely one of the annoyances now. My workaround for now is hovering my mouse over the control bar. For whatever reason that removes the artif
76.
by emcq 10y ago
By definition an optimization being premature means it's not necessary in the present. For these situations its like taking a loan for technical debt. In the future you may need 10x the resources to fix but in many environments that&#x
77.
by emcq 10y ago
If you have good features there is little advantage to a complex model. In production ML there are still many applications for random forests, linear models or svms. Though I prefer random forests because they require less preprocessing, ar
78.
by emcq 10y ago
Perhaps due to the article starting with discussing how AI might be overhyped, but I'm very much not blown away by this post. Reinforcement learning for self improving robots is one of their called out areas? I've never found comp
79.
by emcq 10y ago
They are just describing power iteration which is a standard technique for finding the primary eigenvector. It is commonly used for web scale recommendation problems which likely explains it's usage given the author's prior backgr
80.
by emcq 10y ago
SVN is popular for hardware groups. Why? Better sparse checkout than git when you are dealing with large binary design files. It's not what git is optimized for.
81.
by emcq 10y ago
They had a factory line and were manufacturing devices, but they weren't able to reach their claimed production quality and never shipped. I'm not sure how many failed devices they made which could easily eat millions in material,
82.
by emcq 10y ago
Thats an interesting model, but I cant think of many projects where the features are added at a constant rate. The deluge of pull requests going to Linux are not constant, and probably grows with the number of contributors perhaps relative
83.
by emcq 10y ago
Can you give an example of a project maintained by a large group of people written with monads? The only one that comes to mind is Spark, and while I haven't used it recently, was nightmarishly buggy with many subtle corner cases :&#x2
84.
by emcq 10y ago
I might be confused, but there is ample evidence that you can write maintainable code without monads. Take the Linux kernel, NASA rovers, or redis - all written in boring old C with no monads and great results.
85.
by emcq 10y ago
If you are memory bound the increased throughput from 64 bit to 128 bit wide LPDDR4 bus would likely get you close to 2x.
86.
by emcq 10y ago
This is an exiting architecture! * 2x Denver cores optimized for serial execution (looks like optimized with larger caches and more superscalar speculative execution) * 4x Standard looking A57 cores * 256 Pascal SPs * Balanced increasing co
87.
Nvidia Announces Jetson TX2(devblogs.nvidia.com)
3 points by emcq 10y ago|1 comments
88.
by emcq 10y ago
I've never tried pressure cooking my eggs, but in minimizing prep my favorite way is to steam them. Put a little water in a pot (maybe half an inch), cover, bring to a boil, then reduce to low for 7 minutes and you have a perfect hardb
89.
by emcq 10y ago
I'm not sure what the original post is referring to but many early GPUs were not fully IEEE FP32 compliant, where many operations (particularly trig functions if I recall) would have 24 or fewer bits, either not supporting FP32 or requ
90.
by emcq 10y ago
This is a great point. However, I would generalize a bit and say the problem is sometimes beyond ego; when someone's technical aptitude is greater than their ability to communicate. This is a more common problem than just the smartest

More ›