Posts by Bronek Kozicki

I just happen to work with TensorFlow, PyTorch and Horovod at this moment. These are written in C++ but meant for consumption in Python. That is, the Python bindings are part of the interface around which these projects are designed, from the inception. And it makes sense, given that users of these libraries are data scientists who are unlikely to know C++ but are likely to want to use many other Python libraries in the same project.

It just sometimes makes sense to use, and write in something else than C++ - even if it happens to be ones most preferred language. It is actually perfectly normal that the language is bound to domain, and not all domains have affinity to C++ (or C).

Having said that, I do agree that a VM for a language (JVM CLR and beam) is a very annoying waste of CPU cycles. Sadly we are not going to get rid of them in the foreseeable future, so might as well learn.

Re: It's not us it's THEM..

Well, I don't know. It could be that the Chinese already have all the relevant know-how and do not need to spy on European or American competitors. Or, you know, they just might want to make use of the legal immunity, implicitly provided by the local government, and have a look at what their competitors are doing, of course only to strengthen the domestic market and nothing to do with the actual spying, which as we know is evil.

Re: Device drivers

You put a good argument here, but I disagree. First, the WSL1 shim is not very good as demonstrated by the huge overhead of the filesystem operations and incomplete implementation. Secondly, it was stuck at the emulation of the old kernel 4.4 and no strategy to bring the version forward. They figured it will be more cost effective to use the real thing, with some GPL-ed patches added and I think that's the right decision (even though I am unlikely to by affected by it). Hypervisor technology has progressed significantly in the few past years: all cloud providers rely on it, you can now have hypervisor on your phone without realizing it, and running a home computer inside a VM for AAA gaming is perfectly normal (for some).

Just get rid of the notion that what VMWare is selling to the world is somehow "top of the class" because it is not. You might also investigate the IOMMU device pass-through while learning about more modern VM.

I'd be fascinated to learn more about it, show me some deep neural networks using the approach you are advocating.

Re: I'm confused

Exactly this, IMO the firm has earned respect for demonstrating understanding of the security and willingness to disclose the incident.

Re: Timzones

I missed a dentist appointment many years ago because of exactly such "mis"-feature. The clincher is that this has happened with an old Motorola Razr dumb-phone.

Sadly, that's not how US police works.

Re: My latest batch of...

All-in-all, those little notes are excellent sources of amusement.

they are quite useful to me - for tuning the selection of RBLs to use on a small postfix I happen to administer. Just today added barracudacentral for the one which has slipped by spamcop, abuseat and spamhaus.

Re: 16 TB. And if it fails?

Otherwise just do it like everybody else, acquire a big enough tape library for things which must be backed up and experience data loss on things which do not. For a small churn of data, send deltas to offsite or to the cloud.

You keep using that word "burnout", but I don't think you know what it means.

Re: I prefer

... it is a common problem :(

... forgot to add: but without Gnome.

Re: Seems pretty good

Given that only 417 scans were used for training, I'd say it is very good result.

Yes, especially since the first sentence is closely followed by:

"it’s the cost of building enough infrastructure to overcome distance and therefore reduce latency to make streaming scalable and viable as a business that is the limiting factor."

But I wouldn't call him an idiot. He obviously realises there is a problem with the economic model where the number of GPUs in the datacentre is in linear relationship to the number of users outside it. The idea of using a special protocol for sending drawing data on the higher abstraction level, so that no rendering is performed on the server is certainly not new, but in the context of AAA games it does appear to be a novel solution and I wish him luck (he will need it).

Re: Here we go...

I think you meant 451 http error, 403.

Re: So much potential

It so happens that the only byproduct of burning hydrogen in oxygen is H2O. Taking the space launch industry in this direction would definitely help reducing the CO2 impact it has on the atmosphere.

Re: Sale of the Century

decided to shoot itself in the foot and move away from leading database product

nah, I suspect that the real engineers who deal with large distributed systems at Amazon got tired of Oracle's inability to provide so simple data consistency guarantee like actual serializable isolation.

... and there is also a question of weight.

Re: <Evil Grin>

been through that. Actually, I add coffee cold turkey (for one day only!) to the jet lag, and the relief on the second day is good enough to kill the jet lag much faster than it would normally go.

Re: Rather useless without a different operating system approach

Well, I guess you want your file system to be random access, with allocation before write, release when done etc. Do lookup NVDIMM.

Re: "No one really understands why machine-learning code is so brittle"

Another good point. A well-presented ML will typically return a prediction, which is interpreted as probability value. If the value is exactly 0 or 1 then either the model is broken or rounding error got in the way. Typically, for a great quality model and where the match is perfect, the value might be up to 0.97, perhaps 0.98. It is up to humans to actually read this value and think "hmm, this could be something else with 3% probability".

But then, humans just love survivor bias : if the ML was right more than 10 times in a row we stop paying attention and replace thinking with generalisation. It is good thing that some researchers are actually pushing the probability value closer to 1, but we also need to start paying attention. Because it will never be 1 and yet it will be frequently interpreted as such.

Re: why on Earth give this away for free to everyone on the planet

Actually I think "Perhaps the NSA's enemies are assumed to have better or similar tools" pretty much this.

Actually, we are looking at the collision of two asteroids, and the bits which coalesce are the larger asteroid pulling itself back together (literally, by gravitational force) after the collision. Which means that if we tried to break apart a large asteroid on a collision course with Earth, this is likely to happen with the asteroid, if left for sufficient time.

Re: It's epidemic

They are normal in the UK bank sector :-( And they will stay normal until the regulators force banks to improve, rapidly.

No, really.

Basically what we are seeing is the inability of bank management to manage the technological risks properly.

Re: @Someone Else. A Question:

@Ian Michael Gumby

Either you do not understand how the law works in large parts of the world (including the US) or I somehow missed that this decision would not set a precedent. In case this is the latter, please kindly show the source.

In case there is a precedent, the decision basically means that whoever first grabs copyright for a function name and signature will be able to prevent everyone else from using the same name and signature in their own code. Which is very bad news for developers, because there are only so many sensible names and signatures (as well as code idioms and design patterns) which we can use to solve a common reappearing problem. It also makes it very difficult for developers to change jobs or contribute to open source projects because anything we have done in the past becomes huge legal liability.

Re: Infrastructure

One might have hoped that principles of chaos engineering were applied, to ensure soft fail and quick recovery. One would be wrong.