Start Up No.1,005: Google backtracks on Chrome adblock block, make your own cloud, Apple hires to up IoT game, how AI is messing up science, and more

Brexit effects mean the UK’s House of Commons needs four sides rather than two, according to a new study. CC-licensed photo by UK Parliament on Flickr.

You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.

A selection of 10 links for you. Interactive, but not graphically. I’m @charlesarthur on Twitter. Observations and links welcome.

Google backtracks on Chrome modifications that would have crippled ad blockers • ZDNet

Catalin Cimpanu:


At the root of Ghostery’s benchmark into ad blocker performance stands Manifest V3, a new standard for developing Chrome extensions that Google announced last October.

The long-winded document contained many new rules about what Chrome functions and APIs an extension should use. One of the modifications was for extensions that needed to intercept and work with network requests. Google wanted extension developers to use the new DeclarativeNetRequest API instead of the older webRequest API.

This new API came with limitations that put a muzzle on the number of network requests an extension could access. It took some time before ad blocker developers caught on to what this meant, but when they did, all hell broke loose, with both extension developers and regular users accusing the browser maker of trying to kill third-party ad blockers for the benefit of Chrome’s new built-in ad blocker (which wouldn’t be impacted).

Chrome engineers justified the change by citing the performance impact of not having a maximum value for the number of network requests an extension could access.

But the Ghostery team disagreed with this assessment.

“This work [referring to the study] was motivated by one of the claims formulated in the Manifest V3 proposal of the Chromium project: ‘the extension then performs arbitrary (and potentially very slow) JavaScript’, talking about content-blockers’ ability to process all network requests,” said Cliqz, the company behind the Ghostery ad blocker.

“From the measurements, we do not think this claim holds, as all popular content-blockers are already very efficient and should not incur any noticeable slow-down for users,” they added.


Basically, it seems Google wants to stop any adblockers that aren’t its own, because it wants the choice of which ads are blocked to be its own, not users’.
link to this extract

How Brexit has created four new political factions – interactive graphic • The Guardian

Josh Holder:


Our study clusters MPs by the similarity of their voting patterns: if two MPs always vote the same way, the chart groups them tightly together.

The patterns on key Brexit votes reveal the emergence of four cross-party political factions that are wrangling for control of the negotiations.

A cross-party group of pro-European MPs usually votes with each other, with or against their own frontbenches, while Europhobe Conservatives now constitute a party within the party.


The votes described here probably won’t mean anything to anyone outside the UK, but scroll down and the evolution from two-party system to four-group dynamic becomes clear. The Europhobes are indeed a party within a party in the Tories, and extremists outside the party in Labour. The Europhiles, meanwhile, are effectively homeless, politically. There are rumours that the Labour Europhiles will break away this week.
link to this extract

Dropgangs, or the future of darknet markets • Opaque Link


To prevent theft by the distribution layer, the sales layer randomly tests dead drops by tasking different members of the distribution layer with picking up product from a dead drop and hiding it somewhere else, after verification of the contents. Usually each unit of product is tagged with a piece of paper containing a unique secret word which is used to prove to the sales layer that a dead drop was found. Members of the distribution layer have to post security – in the form of cryptocurrency – to the sales layer, and they lose part of that security with every dead drop that fails the testing, and with every dead drop they failed to test. So far, no reports of using violence to ensure performance of members of these structures has become known.

This concept of using messaging, cryptocurrency and dead drops even within the merchant structure allows for the members within each layer being completely isolated from each other, and not knowing anything about higher layers at all. There is no trace to follow if a distribution layer member is captured while servicing a dead drop. He will often not even be distinguishable from a regular customer. This makes these structures extremely secure against infiltration, takeover and capture. They are inherently resilient.

Furthermore the members of the sales layer often employ advanced physical tradecraft to prevent surveillance by the procurement layer when they pick up product. This makes it very hard to dismantle such a structure from the top.

If members of such a structure are captured they usually have no critical information to share, no information about persons, places, times of meeting. No interaction that would make this information necessary ever takes place.

It is because of the use of dead drops and hierarchical structures that we call this kind of organization a Dropgang.


We ain’t on the Silk Road any more.
link to this extract

Differential privacy: an easy case • Substack

Mark Hansen:


By law, the Census Bureau is required to keep our responses to its questionnaires confidential. And so, over decades, it has applied several “disclosure avoidance” techniques when it publishes data — these have been meticulously catalogued by Laura McKenna, going back to the 1970 census.

But for 2020, the bureau will instead release its data tables using a “formal privacy” framework known as “differential privacy.”

A unique feature of this new approach is that it explicitly quantifies privacy loss and provides mathematically provable privacy guarantees for those whose data are published as part of the bureau’s tables. 

Differential privacy is simply a mathematical definition of privacy. While there are legal and ethical standards for protecting our personal data, differential privacy is specifically designed to address the risks we face in a world of “big data” and “big computation.”

Given its mathematical origins, discussions of differential privacy can become technical very quickly.


Apple and Google use this to make it harder to de-anonymise personal data. This is quite a long post, but it explains it while sticking to quite simple maths.
link to this extract

Internet censorship: Facebook, Patreon will always be frustrating • Bloomberg

Tyler Cowen:


Facebook recently has devoted a lot of resources to regulating speech on its platform. Yet undesired uses of the platform hardly have gone away, especially outside the U.S. Furthermore, the need for human judgment makes algorithms increasingly costly and hard to scale. As Facebook grows bigger and reaches across more regions and languages, it becomes harder to find the humans who can apply what Facebook considers to be the proper standards. 1

I’d like to suggest a simple trilemma. When it comes to private platforms and speech regulation, you can choose two of three: scalability, effectiveness and consistency. You cannot have all three. Furthermore, this trilemma suggests that we — whether as users, citizens or indeed managers of the platforms themselves — won’t ever be happy with how speech is regulated on the internet.

One view, which may appear cynical, is that the platforms are worth having, so they should appease us by at least trying to regulate effectively, even though both of us know they won’t really succeed. Circa 2019, I don’t see a better solution. Another view is that we’d be better off with how things were a few years ago, when platform regulation of speech was not such a big issue. After all, we Americans don’t flip out when we learn that Amazon sells copies of “Mein Kampf.”

The problem is that once you learn about what you can’t have — speech regulation that is scalable, consistent and hostile to bad agents — it is hard to get used to that fact. Going forward, we’re likely to see platform companies trying harder and harder, and their critics getting louder and louder.


(Via Nathan Taylor’s fine roundup.)
link to this extract

The cloud is just someone else’s computer… but what if it were your computer? • Coding Horror


Given the prevalence and maturity of cloud providers, it’s even a little controversial these days to colocate actual servers, but we’ve also experimented with colocating mini-pcs in various hosting roles. I’m still curious why there isn’t more of a cottage industry for colocating mini PCs. Because … I think there should be.

I originally wrote about the scooter computers we added to our Discourse infrastructure in 2016, plus my own colocation experiment that ran concurrently. Over the last three years of both experiments, I’ve concluded that these little boxes are plenty reliable, with one role specific caveat that I’ll explain in the comments. I remain an unabashed fan of mini-PC colocation. I like it so much I put together a new 2019 iteration…

…Let’s break this down and see what the actual costs of colocating a Mini-PC are versus the cloud. Let’s assume a useful life of say, three years? Given the plateauing of CPU speeds, I think five years is more realistic, but let’s use the more conservative number to be safe.

$880 mini-pc 32GB RAM, 6 CPUs, 500GB SSD
$120 taxes / shipping / misc
$29 × 12 × 3 = $1,044
That’s $2,044 for three years of hosting. How can we do on Digital Ocean? Per their current pricing page:

32GB RAM, 8 vCPUs, 640GB SSD
$160 × 12 × 3 = $5,760


Colocation is quite a thing.

link to this extract

AAAS: Machine learning ‘causing science crisis’ • BBC News


Machine-learning techniques used by thousands of scientists to analyse data are producing results that are misleading and often completely wrong.

Dr Genevera Allen from Rice University in Houston said that the increased use of such systems was contributing to a “crisis in science”.

She warned scientists that if they didn’t improve their techniques they would be wasting both time and money. Her research was presented at the American Association for the Advancement of Science in Washington.

A growing amount of scientific research involves using machine learning software to analyse data that has already been collected. This happens across many subject areas ranging from biomedical research to astronomy. The data sets are very large and expensive.

But, according to Dr Allen, the answers they come up with are likely to be inaccurate or wrong because the software is identifying patterns that exist only in that data set and not the real world.

“Often these studies are not found out to be inaccurate until there’s another real big dataset that someone applies these techniques to and says ‘oh my goodness, the results of these two studies don’t overlap‘,” she said.

“There is general recognition of a reproducibility crisis in science right now. I would venture to argue that a huge part of that does come from the use of machine learning techniques in science.”


Reproducibility means what it says: can you get the same result starting from the same data? This isn’t a new problem – it’s just becoming more visible.
link to this extract

The convergence of the phone and laptop • AVC

Fred Wilson:


The Gotham Gal [Wilson’s wife, who also works in venture capital] wanted to get a new laptop. Her late 2015 Macbook has started to fade on her.

So yesterday we made a visit to the local Apple Store and checked out the options. We looked at the Macbooks, the Macbook Airs, and we also looked at the iPad Pros. We debated the choice and she ended up deciding to go for the iPad Pro. We work with a few people who have iPad Pros and love them. And she noticed how much I am using and enjoying my Pixel Slate.

One of the most interesting things about these hybrid tablet/laptop devices is that they run operating systems that are designed for the tablet or phone. They are touch devices like our phones vs mouse devices like our laptops.

A good example of this is how I do email on my Pixel Slate. I could run Gmail in the browser on my Pixel Slate. But I have found it much more pleasing to do email in the Gmail Android App on my Pixel Slate. I swipe emails away like I do on my phone. But I also have the keyboard when I want to write a long response. It is literally the best of both worlds.


I think she’s going to be happy with it, though I wonder what is actually meant by saying a 2015 machine “has started to fade”. The comments on the piece are worth reading too: as many saying she’ll go back to a laptop as saying the tablet is the way forward.
link to this extract

Sam Jadallah to join Apple as its new leader on smart home devices • CNBC

Christina Farr:


Jadallah previously ran a start-up called Otto, which made a $700 lock that was backed by the venture firm Greylock. He also spent more than a decade at Microsoft, and had a stint in venture capital at the firm Mohr Davidow.

Otto suspended its operations four months after launching its beautifully-designed Bluetooth- and Wi-Fi-enabled luxury lock. In interviews, Jadallah hinted at having found a buyer, which pulled out at the last minute.

About 70% of the early team behind Otto were actually poached from Apple’s ranks, Jadallah has previously said. The lock was compared favorably by reviewers to the “Apple of smart locks.” It’s not clear whether Jadallah will bring these early employees with him, or will have a fresh mandate to hire. There are currently about half-a-dozen job openings in Apple’s home division.

But Apple also competes against rivals like Alphabet and Amazon, both of which have had a head start on moving into the home.


As was discussed on the latest Talk Show podcast by John Gruber and Rich Mogull, it makes no sense for Apple to offer an Apple TV, HomePod, integration to home devices via HomeKit, and yet not have something comparable to Eero or the Google Home router. Maybe Jadallah will see that too.
link to this extract

Major games publishers are feeling the impact of peaking attention • MIDiA Research

Karol Severin:


Earlier in February, Electronic Arts (EA) reported disappointing quarterly results; now Activision has laid off nearly 800 staff, mostly in marketing and sales.

As MIDiA has reported multiple times before, engagement has declined throughout the sector, suggesting that the attention economy has peaked. Consumers simply do not have any more free time to allocate to new attention seeking digital entertainment propositions, which means they have to start prioritising between them.

This downward trend in engagement has persisted for a while now, and the latest quarterly results from some major games publishers confirm that a revenue slowdown will ultimately follow consumer behaviour. Arguably sooner than most of the games industry would have thought.

Publishers will be quick to blame declining engagement and revenues on Fortnite. While the title indeed intensified the manifestation of the peak attention economy dynamics among gamers, the coming slowdown is part of a much bigger challenge – how to capture attention in an increasingly attention-scarce landscape…

…Not only is engagement declining across mobile, PC and console gaming, at the same time, video is winning the race against gaming in capturing attention on multipurpose devices such as PC. Between Q1 and Q4 2018, gaming on PC declined faster than TV show viewing on PC among gamers.


They’re all watching Twitch or YouTube rather than playing the games themselves. It’s like football losing its park players. But the idea that the attention economy can’t grow any further is quite dramatic.
link to this extract

Errata, corrigenda and ai no corrida: none notified

2 thoughts on “Start Up No.1,005: Google backtracks on Chrome adblock block, make your own cloud, Apple hires to up IoT game, how AI is messing up science, and more

  1. re. Google adblock. I think the crippling of adblockers was just a side effect. Google has been working on a lot of “enhancements” to make sure the web is fast, some of them with controversial side effects (AMP, SPDY, Never Slow, and that reining-in addons thing).

    There’s an obvious conflict of interest especially since tracking and ads are the worst cause of slowdowns. I think Google has been relatively ethical in not favoring its own ad platform, but that’s very much a grey area. Google-ad has if not input at least an inside track into what’s coming to Chrome.

    This is were openness serves as powerful check and balance. Even on Android, you can switch browsers, so if Chrome gets too bad, it’ll just get swapped out. I already did that in favor of Firefox which supports full addons on Android. And this is why I’m still not on ChromeOS, though nowadays I could run Firefox on Android within CHromeOS… sound a bit convoluted.

    I’m a bit puzzled by the focus on performance. Maybe I’m a pushover but I don’t mind waiting for stuff for 5 seconds, and when it takes longer than that it means the network connection is bad and I won’t be able to do much anyway. I’m curious if things are different elsewhere, or if it’s more about bandwidth and not clogging up pipes rather than actual speed ?

  2. re. Machine Learning: I’m no expert, but it noticeably has the same issues as maths in general and statistics in particular:
    – know you formulas and how to play with them
    – know when to apply them
    – Know when NOT to apply them: double-, triple- and quadruple-check you’re not Doing It Wrong.

    At least for me, that last line is the hardest. One can get lost in playing with data and utterly lose track of what it represents and means. AI can by definition find structure and logic in anything. I’m surprised there’s not a standard for AI audits: you should spend 10% of the effort of doing something on evaluating what you’re doing. AI doesn’t have that independent second-guessing/auditing mechanics. I see it daily with nonsensical voice dictation results that produce sentences no-one ever used that are very adjacent to very common sentences. We have AI, now we need Sanity Check. Across the board.

    (BTW, my iOS & Android equilibrium calculation was iffy, but I chose to look the other way… I’m curious what happens if the time series are divergent, it’s luck they aren’t)

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.