AWS bills by the second, and AI reaches it limits on security

Mostly because AI used for committing crimes can be better targeted and doesn’t really care about being wrong (see, for example, this story about password-guessing models), and because AI security software presently creates too many false positives. Here advice: hire more security people and have them patch like crazy. 

If you’re reading this, I assume you’re hip to the fact that Canada—and western Canada, in particular—has an abundance of AI talent in its universities. Facebook joins Microsoft and Google (at the least) in establishing a presence in the Montreal area, although Google (via DeepMind) has also set up a lab in Edmonton. Also, this picture is worth a thousand words if you’re into deep learning gurus and world leaders.

Matroid is a cool company tackling what has to be a pretty widespread issue: reams of video (including surveillance video) without anyone knowing what’s in it. As a consumer, I would love more intelligence and object recognition in products like Nest, Ring, Canary, etc.

On the one hand, this is a really cool little story from Microsoft because F1 racing (from what I understand) values technology and data analysis almost as much as it values good driving. On the other hand, Microsoft’s partner, Renault, isn’t exactly lighting up the circuit.

And, in this case, investors bought into it and raised Nvidia’s share price by 6 percent. But let’s be honest: Nvidia’s long-term lead in AI is hardly a given considering (1) the big difference between the number of machines training AI models vs. the number of devices (e.g., smartphones) running the algorithms, and (2) the insistence by cloud providers on building their own custom AI chips.

It does seem like there’s an uneasy relationship between AI and neuroscience, because we don’t understand entirely how the brain works. The consensus seems to be AI research should use the brain for inspiration, but stop trying to copy it because it might be limiting thinking about how AI should really work. In related news, lots of folks are still working on brain-inspired chips and Google’s Geoff Hinton thinks AI researchers might need to retire a popular technique because it’s not how the brain works.

This is a little marketing-y, but I think does a good job laying out how the pieces to do some meaningful work with AI are already in place. I don’t think companies necessarily need an AI strategy, but they should at least be thinking of ways they can utilize it, even for simple things to start.

Yes, and no. People who are going to do unethical stuff will do it with or without AI, although AI can make some things easier and/or provide an air of legitimacy in certain situations. If anything, I think AI will just require citizens, critics, journalists and other people to ask more critical questions and not take research findings, product claims, or important algorithms at face value.

This is potentially important work as we try to figure out the right balance on human-computer interaction—in this case with regard to training autonomous vehicles in real time.

I’m trying to decide how big of a deal this is, because I don’t think of compute or storage instances being as ephemeral as containers or lambda tasks. But, still, the pricing model is definitely headed in the right direction in terms of granularity and making shorter-lived workloads more affordable. Can we safely credit Google for making pricing models a point of competition when it introduced per-minute billing, sustained use discounts, etc.?

So that makes everybody, right? Also, I think AWS is still the only cloud provider rolling the semi up to your data center.

Because while things like cloud and continuous delivery can greatly speed up the development process, they also open up that many more opportunities for misconfigurations.

Dropbox is (relatively) silently growing into a large company that needs to handle some serious traffic and performance requirements, which typically does require moving away from public cloud services. This is a very informative post on how the company overhauled its backbone to drastically increase bandwidth and decrease latency.

Prediction: Assuming Oracle delivers this database in the near future, it will give the company an advantage for about a minute. Cloud providers all see Oracle’s business as something from which they’d like to take a big bite, and they are no strangers to automation.

In other Oracle news, it’s rolling out new SPARC chips just a short while after reportedly laying off a goodly chunk of its hardware division.

Apparently, the ever-quotable Chamath Palihapitiya of Social Capital made a comment about how processor improvements will render large data centers all but moot, and the result was a slide in data center stocks. I take the side of this argument, which is that large data centers will continue to be a thing for large cloud providers and web companies—better processors just means more work per square foot, but there’s always more you can do. Plus, the advent of hybrid clouds and edge computing means even smaller companies will want broader footprints co-located with network access.

In related news, data center provider Switch, which is headquarted a few miles down the road from me in Las Vegas, is growing globally and is now set to IPO.

This is definitely worth reading if you’re into design patterns for microservices and functions/lambdas/serverless. But there’s also the bigger picture, which is that we have to take Uber seriously as an engineering shop even while turmoil surrounds its business.

An AI algorithm that can cut data center cooling costs by 10 percent would have a lot of takers. I believe Google has applied something similar internally, as well. I have to assume somebody is trying to monetize on something like this as we speak. 

I hadn’t heard of Panopoly, but now I have and you have, too. From what I can tell, it’s selling a technology that uses machine learning to try and automatically uncover insights in data. 

In this case, its Jay Kreps arguing for Kafka as a data store, which is in part related to the company’s recent announcement of a Kafka-based SQL dialect called KSQL. If the conventional wisdom in enterprise IT and open source is that you want to be a platform rather than just a component, Confluent is doing its damnedest to help Kafka transition from the latter into the former.

The Transaction Processing Performance Council, which has been benchmarking databases for years, has set its sights on IoT data. Say what you will about benchmarks, but it’s probably good to get any standards in place for enterprise IoT, even if it’s just testing ingestion rates of the gear sucking up all that device data.

Pandas, if you’re unfamiliar, is a popular Python-based data science library. And, according to its creator, it’s also being taken advantage of by large users that don’t contribute anything back to its development. Beside the threat of something happening to a widely used project is the still-unsolved question of open source licensing.

The aptly named Query Analyzer. This is an interesting blog post if you run a lot of MySQL jobs, but I think it will be much more interesting once the project is open sourced.

Metamarkets has been on a tear lately with insightful engineering blog posts, and this one about how it bin-packs Spark and Druid jobs on its Mesos cluster is no exception. That being said, I also read today a detailed piece on data analytics at Russian search engine giant Yandex, which built and open sourced its own Druid-like system called ClickHouse and counts CloudFlare among its users.