Recently the French tried to set up some digital taxes for these IT megacorporations (Facebook, Google, etc). Trump threatened to put taxes on a load of French products in response. Yet his party is making sure these corporations don't even pay taxes in the US. Just a naive question, whose side are the US politicians on? ...

This seems most likely. It's not they are banning a ban on Huawei in the US. But they don't want to stop providing Huawei with backdoored products.

European shoe size or US shoe size?

code always seems to become a mess. especially in a project with a few developers, and i would guess an indie game is a great example. someone who claims their personal project a mess likely is also more critical of themselves anyway.

never heard of this game but it is still nice to see source code released for published games. the game is also still sold on steam, has 5k+ reviews and very positive rating. so i guess some people have heard of it.

the blog post goes very nicely with the code, providing interesting story elements on its development besides the single line picked up in this slashdot summary stating the code is a mess.

the mess seems to refer to some basic controversies of code quality (e.g., passing everything around all the time because "globals bad"), just hacking it together fast, and stating how he would be slower now trying to do it "right". is it so bad to just hack working stuff together while having some fun when making a simple indie game?

I often wonder what happens to "old" software engineers. Their numbers seem to keep halving the older they get. I have seen some become sheep farmers, cosmetologists, or whatever. Interestingly this person did not go crazy in the end but rather someone to help us all after doing too much software engineering. A psychotherapist. Maybe he could specialize in treating burnt out software engineers. Good experience.

Didn't Donald Trump, in his great and unmatched wisdom, just announce the United States Space Forces?

This is obviously an opportunity to use the space debris for target practice and space-weapons development. For cleaning up the environment, what else do you need? How could anyone oppose?

The label "nothing" seems to be just the author referring to his example of an empty box containing nothing, something people learn to recognize as a concept of "zero" items in a box.

What he actually seems to be talking about is identifying anomalies, or instances that seem to greatly deviate from previously seen instances. The author being a CEO of an "AI-powered visual inspection company". So an example of his problem is to find defective items (e.g. scracthed) on an assembly line.

With very few instances of the "bad" (e.g., scratched) class, he just wants to find items that deviate from the norm (scratched). And from this comes the idea of a class label for "nothing", meaning the item in question seems to deviate much from anything the system was trained on. Or maybe it just deviates a little if scratched... Seems you would need to tune quite a bit to the problem as well.

It is an interesting thought more generally. Why only train with apples and bananas, when you could throw a large library of training images of other types of items on it, label them as "not what I am looking for". Haven't really seen much research into how to do that effectively. Or practical examples. Would be interesting.

Obviously there is a need to process the big IoT data, machine learn from it, and finally turn it into an AI system to control it all.

It's an interesting thought. Various banks holding their own blockchain records, adding blocks, ... How do the current international banking records actually work? There must be some way to achieve some confidence that the other party did not just make up the "money". OK, banks "make up" money all the time, that is how the financial system works, but there must be some ways to have shared trust already. What are they?

If banks were the trusted parties in a blockchain, then how would two customers change funds? Need to send the transactions to a bank of choice to be included? If they have different banks, how long would it take for their chains to sync the transactions? Or is it just a single chain? How are the blocks verified across? Hold a key ceremony (swap party time) and all "banks" exchange keys they later use to sign the transactions and blocks? Give each customer a key to use? Obviously you don't need miners to hash stuff, just signed by the parties? How is this different from the current system again?

It seems this could be made to work. But I have not yet seen a working system, or an explanation of one of this kind.. One that would explain the special benefit of blockchains in this, and their exact role that makes them special, or the overall resulting system itself.

With Bitcoin, XMR, and various "shitcoins", it is quite obvious, since the idea is to decentralize and avoid the banks. Since avoiding the banks is the ultimate goal they had, this makes the questions even more interesting..

After years on the hype-train, they are finally getting closer to providing pre-trained models for more tasks than imagenet image classification. Pre-trained on large-scale datasets, and light-weight versions of the frameworks to deploy on mobile devices. Scaling up, scaling down. Machine learning for developers to apply, as it should be. OK, thanks.

You sir nailed the description of life. And was labeled Troll for it. I guess truth is not popular.

I have tried to look up Hyperledger a few times to understand behind the hype. I never find anything but the hippy marketing material. I get Bitcoin and the way it uses various technologies to build its "decentralized trust". I think it's a super-clever solution that Satoshi came up with. A mashup of existing technologies, but very cool and innovative in my view.

But this whole thing about a private blockchain like Hyperledger. I do not see how is this different from a centralized (distributed?) database. If you don't have people "verifying" the content by hashing it, varying difficulty, distributed storage and "consensus", then what is the point of this "blockchain"? Even in Bitcoin type blockchains the "consensus" is just that network nodes agree that the hashes in the chain match, and the data in the ledger is thus unmodified. Because they share the chain data, and the hashes match. That's the consensus.

Now if you host the blockchain centrally, and the blockchain, as well as the blocks are managed by IBM or whatever company, how is this different from a database? They should be able to rewrite it any way they like. They can control whatever they want to put in there. Burying your data in a pile of blocks, as a chain of metadata hashes could be a useful way to authenticate the inputs to news articles. But what does the private / permissioned blockchain give you that a database wouldn't? Are you going to share the blockchain all the time but only let some people write to it? You can do that with a database.. As can you hash the contents, and whatever.

So, anyone, what is the point of a private / permissioned blockchain? Just a trendy name for a specific database technology and nothing really to do with the origins of the term in Bitcoin etc?

I am sure there are many big systems and organizations where this change is a huge undertaking. And management keeps pushing it forward because, much like testing, it's not a feature you can sell. Until brown substance hits the fan, at which point no-one remembers these discussions, of course.

Then there are the people who don't want to change. At the office I have promoted moving on to Python 3 for more than a year I have been there, and people keep coming up with responses like "but Python 2 will still run". Pointing out new features or improvements it would allow with much less effort is meaningless, because changing from Python 2 to 3 is too much of a mindshift. Not that they tried. I have looked at the code and ported parts of it as an experiment, and it is really quite trivial changes. But even installing Python 3 for some is too much, because "it might break my dev environment". Good luck ever changing anything with that.

So I guess it's a complicated problem. Personally I am solving this problem (and various others it is a symptom of) with the help of the job market..

Maybe AWS spends enough on HW to justify spending on their own designs.

Related to this, I was a bit surprised to notice Gosling is also working at Amazon these days on latest versions of the JVM for these Gravitrons.

https://aws.amazon.com/blogs/c...

Having modern SW platforms available on the Gravitrons should make running code simpler. So any JVM language and whatever else you can support in a similar way, I guess (Python etc?).

It's seems to me this is just some random persons blog post about their personal experiments on learning and trying this stuff. Interesting to see, even if nothing very advanced or special.

I tried to play a bit with this type of data in a recent Kaggle competition. Using LSTM's, gradient boosted trees, etc. News headlines, social media data, price history, ... It all gets quite complicated. Although I suppose you don't really have to beat the market by more than a small margin and then make a good trading strategy. If only someone was foolish enough to post a successfull approach on this to learn from.. :)

(probably such examples exist, just not aware of them, not looked too much)