Dear YouTube

Dear YouTube,

You have done so much to enable creators to make and share exciting videos, whether that be mini series, cooking shows, films, music videos, news, podcasts, top 10s, documentaries, you name it.

Many communities have spawned around channels and topics, with comment sections ranging from support, to fully-blown drama. And whilst there is quite a lot of trolling, the positive and supportive comments far outweigh the negativity.

However, many of us viewers are frustrated. Those of us who spend hours watching content (and ads) have been receiving videos with worse content during the past year, more clickbait, less meat, more in-video ads, and fillers due to the 10 minute mark for mid-roll ads.

Additionally, too many creators are making videos complaining about YouTube’s policies and demonetisation, even on channels that didn’t use to post regularly about that (e.g. PewDiePie, h3h3Casey Neistat, Philip DeFranco, MrBeast, Jörg Sprave, and a myriad of other creators). A quick search for “demonetized” on YouTube returns around 369,000 videos. That’s a lot of angry creators.

This means that these YouTubers are derailing from creating the content they are known and loved for, and instead they resort to making videos talking about the platform’s problems that are directly affecting them, and in turn this means we, as viewers, receive less of the content we actually want to watch.

The way we see it is that there are three groups of problems currently plaguing the platform: monetisation, content rules/flagging, and UX issues.

1. Problems regarding monetisation:

After demonetisation was introduced over a year ago, a lot of creators were concerned, and after the first few high-profile cases started coming in, users and news outlets started calling it the “Adpocalypse“.

One solution that is increasingly being used by creators is to use Patreon, and in some cases that works really well, for example: musicians, artists, writers, cartoonists, animators, etc. But it is still an extra action (and account) a viewer has to make on a 3rd party website, which means a lot of potential contributors will be alienated. This would be solved with channel-based subscriptions similar to Twitch. We know you are experimenting with (and currently rolling out) a paid subscription system for channels, which is a good solution on paper, but in our opinion this is not the best solution.

An option to consider instead might be to establish a pot for monthly payouts. For example, a user might decide to contribute $5, $10, or $20 per month, with a similar model to Spotify or Netflix. The amount then gets divided at the end of the month between YouTubers with a percentage-per-creator calculated by an algorithm based on subscriptions, likes/dislikes, time viewed per channel, amongst other parameters.

In return, viewers are rewarded with fewer ads depending on how much they contribute per month. Google naturally takes a cut from these proceeds.

Furthermore, this can work without disrupting the business model of YouTube Red (now apparently called YouTube Premium). If we contribute $25 or more per month, why not offer the option to subscribe to YouTube Premium, but instead of keeping the $10, your revenue would depend on viewing habits, keeping $5-10 a month depending on the amount of time viewing regular YouTube, versus YouTubePremium exclusives.

2. Problems regarding content rules and bots:

One of the issues here has been communication in the past regarding changes. Recently Susan Wojcicki has stated you will be making changes in 2018 regarding how you communicate with creators. However, the wording seems to favour giving a “heads up” instead of proactively holding discussion of changes that will create major repercussions for creators.

However, that is only one part of the problem. One common complaint is that content decisions regarding monetisation, and the automatic “category / subcategory” the video falls under is not transparent enough. Some people even figured out how to enumerate the category IDs for sensitive content. Interestingly enough, that document also dives into the case of systematically suppressing demonetised videos from “suggested videos”. By making this information visible to creators, they will be able to make decisions that will influence the direction in which they wish to take their channel.

Rules also have to be made explicit, and reasons for flagging, or why videos were categorised in a certain way should be easily viewable in the Creator Studio, along with any additional information that could help the creator understand where he went wrong and how to improve / fix it. Timestamps of infringing content (the same as those applied to copyrighted content) are also a big plus.

Dude Perfect and PewDiePie’s subscribers have also mentioned a couple of instances where X-rated ads were shown on some of his videos, as their videos were placed in categories meant for mature audiences, whereas clearly their viewers have a wide age range.

3. UX and notification issues:

Videos we have viewed before (or almost finished watching and then switched to another video) constantly reappear in the suggestions. These should be excluded by the suggestions filter as few people would want to watch the same video twice (music videos being an exception). And please, do not rework the chronological subscription feed, fix it instead.

The bell / subscribe system is broken and has been mentioned by both viewers and creators. You have denied this vehemently in the past. However, recently a video on one of your own channels you mentioned the issue happened to you and mentioned this would be escalated. In this specific issue it might just have been an MQ (message queue) issue with sending out announcements (or your email service provider experienced downtime).

A mitigation strategy here would be to implement better monitoring of users that requested notifications VS users that received notifications and emails. Most push notification services, as well as email services provide receipts of reception, and also give statistics on the percentage of emails marked as spam. You should graph this (using DataDog for example) and make it visible on a monitor at all times in the HQ; this will allow you to see the actual problem, and formulate strategies to improve notifications.

As a UX bonus, if a creator uploads something new and a user is currently on YouTube, display a toast message mentioning “{CREATOR} just uploaded a new video titled {TITLE}”, you can do this easily with WebSockets or a service such as Pubnub or Pusher. The whole feature would probably only be 20-30 story points (excluding mobile).

What have been the consequences on the current policies and problems?

Jörg Sprave started The YouTubers Union movement recently, where he has 15K+ followers on his Facebook Group. He plans to take direct action such as strikes, and communicate constantly with you to try to convince you to make changes that will improve the lives of content creators.

PewDiePie has been talking about the issues for more than a year, to the point he changed his entire delivery style accordingly. Casey Neistat interviewed YouTube’s head of business (which was a great start, as mentioned in the comments of the video), however, much is left to be done.

As viewers, we don’t really want to meddle with the politics of what is happening at YouTube, nor with the frustrations of the creators. Instead, we want quality content. Many of us don’t watch Netflix, Hulu, Amazon Prime; instead, we rely on Youtube for our daily dose of entertainment. But something is really wrong when a lot of the content we watch includes long segments with complaints or rants about demonetisation, and the other issues mentioned in this article.

Our proposal:

We know the issue of figuring out a compromise between ads, paid subscriptions, and YouTube Premium is a complicated one, potentially involving everyone from your legal team, accounting, engineering, product, designers, and project manager.

An intermediary step may be: While you figure out subscriptions (and please make it easy to pay), then you should at least cover part of the salary of the creators that get millions of views, but are heavily demonetised, maybe based on the amount of views and likes per video.

Afterwards, we would like to offer a few strategies on how to develop and implement the changes. We’ve written these down in the following section.

We would also like to re-iterate the demands posted on Jörg Sprave’s union website, which we have split per section. These give a good idea on some of the major complaints the community of creators has:

  • Monetisation:
    • Monetise everyone: Bring back monetisation for smaller channels
    • Disable the (auto-flagging/demonetising) bots. Add a voting system aided by AI (see Valve’s VACnet)
    • Pay for the views
    • Stop demonetisation as a whole
    • Pay according to delivered value
  • Content / Rules:
    • Transparent content decisions
    • Equal treatment for all partners
    • Clarify the rules

You can read the extended version of the union’s demands on their website.

How can you implement this?

Set up three fireteams, one dedicated to the monetisation issues, each one dedicated to one of the topics mentioned above.

These teams should be nimble and be able to act with few constraints and red tape. I’d suggest the following structure:

  • Each team has 1 senior backend engineer, 1 medior frontend developer, 1 senior frontend developer, 1 QA engineer, and 1 team leader who is also experienced with the code base and can help with code reviews and basic project management. Some teams might require additional backend developers depending on the complexity of the issue.
  • Each team should start the project with a kick-off, determining all milestones from the start. One strategy might be to hold a 1-day meeting with several senior product managers, designers, backend engineers and infrastructure/devops, a compartmentalised list should be brought and each topic should be discussed, either accepted or rejected, and difficulty determined via planning poker.
  • 1 additional project manager who oversees all 3 teams, reports status directly to Susan, and attends daily stand-ups for all 3 teams.
  • 1-2 Infrastructure/devops engineers ensure that CI/CD are setup, test/production clusters are ready, and can immediately help with any needs (subdomains,, anything that needs to be deployed) during the whole part of the process.

This approach, with a positive attitude towards change, and a healthy dose of open-mindedness, may ultimately not only save YouTube whilst feeding content creators, but it may also establish a forward-thinking platform which could outlive us all.

The viewers.

PS. Big props to Dear Github for the inspiration for the style and wording of this post.

PPS. You have our permission to re-upload and modify this post as you see fit.

This post has been reposted on Github where you can make amends via a PR.

How Police could effectively pre-empt riots

Over the past few days we have seen violence escalate dramatically in different cities across England, with London being the primary location. Acts of violence, rioting, looting, and arson have taken place at night for several days.

I have been attached to the BBC News coverage through their website and I saw an interesting tweet posted there:

Tweeted by Hannah Robertson in GloucesterAt first I thought. Pre-empt where the next bout would take place? How on earth would they manage that?

But it actually isn’t that hard. They have a mountain of CCTV data, they have a good amount of servers, all they lack is a bit of artificial intelligence.

The best way I can think of to determine where possible bouts of violence could sprout in what seems like an apparent random non-deterministic method of choice is as follows:

The IT team of the police should annotate the direction of each CCTV camera, including which streets are in view and the coordinates.

Afterwards, the CCTV cameras should be plotted on a 2D map, as vectors (pointing in the direction they are facing, with the length of the vector being the distance the camera covers).

Map Plot of LondonAdditionally, all points of interest should be mapped, such as shopping centres, residential areas, commercial areas, shops with high-value items, and shops with low-value items. Most of this data can be readily obtained from Google Maps, amongst other online maps. This data can be extracted and annotated with the values we require (such as the value of the goods sold per area, etc.)

If you think this is a very hard task, it is not. A very simple way of doing this would be to go to the website (or ask by phone) of each major retailer and chain for a list of addresses of their shops. Google Maps can plot them on a map using a spreadsheet as input. There you go, simple as that!

A Machine Learning program could be developed using WEKA (for instance), in which an SVN is programmed to take data from the map-plot, where priority spots include places where high-value items are sold. Additionally, it would take data from each CCTV camera.

Now, how do we represent the data from CCTV cameras? One way would be to take a selection of pixels from each camera, measure the amount of change for a second, and wherever there is a large enough change in different areas of the same camera, we might have a lot of movement going on. So we automatically annotate the data per CCTV camera as having “movement” or “no movement”. Additionally, a range from 0-1 would produce better results (hopefully).

So now we run an SVN machine on the data and hopefully come up with some interesting results.

What could essentially be obtained from this is a vector describing the movement of as mass of people from camera to camera, this vector would be projected on the 2d map. Multiple vectors could be plotted at the same time if there were multiple riots taking place at the same time in the city. A confidence level can be given to each vector (assuming we have built in a few mechanisms to differentiate people from cars, etc). The places of interest could be mapped as hotspots, and furthermore we can predict paths to possible places of interest, derived from the speed and direction of each vector.

So there you go. If you belong to the MET, please share this blog post with your boss (or the IT team) and get working on a system to perform such actions, it would certainly help prevent further stupidity in the future. 6,000 police officers should be able to deal with outbreaks of crime provided they knew where the rioters would be gathering and where they are heading.

IIS flaw under investigation by Microsoft | Tech News on ZDNet

Microsoft on Monday said it is looking into a report of a flaw in some versions of its Internet Information Services product that could allow an attacker to gain control of a system.

In a statement, a Microsoft representative said the company “is investigating new public claims of a possible vulnerability in IIS 5 and IIS 6 File Transfer Protocol (FTP)”.

Microsoft said it is not aware of any attacks using the vulnerability. “We will take steps to determine how customers can protect themselves, should we confirm the vulnerability.”

According to IDG News Service, code for exploiting the unpatched flaw was posted to the Milw0rm Web site. IDG said the exploit appears to affect primarily older versions of IIS — and only when the FTP function is enabled.

Once it is done with its investigation, Microsoft said, it will decide how to address the matter, which could include a security update as part of its monthly Patch Tuesday or an out-of-cycle update.

via IIS flaw under investigation by Microsoft | Tech News on ZDNet.

Oh this weather…

Heavy storms shattering glass windows, half the city flooded, cars passing by on the nearby river, electricity shortages, several million with no electrical power, quite a few killed, a few drowned, and several electrocuted. Thats what the panorama might have been like if there *only* had been a huracane.

Most businesses closed halfway through Wednesday and the whole of Thursday to prepare for the great storm, which the news blew out of proportion… maybe… the reality is that it just turned away somehow and managed to miss us at Monterrey, NL, México.

Great way to spend a Thursday though, small showers, cool weather, xbox 360, music and more.

Hello World


I’m going to try and write regularly about stuff that comes to my head, lyrics, maybe a few poems once in a while, political thoughts, food for the brain, and whatever else comes to me. Not because I think I’m an interesting person or anything, but because I think we all should have some place to express a few of our thoughts, it sometimes makes life more interesting 😉

Arbitrary text of the day: “This will locate the customer based on the number you entered in the first text box”.