HackerHow much of a preference did the last recommendation create?
A click responds to an item and to the choices that were visible. Follow missing posts and creators adapting to a metric to see how one round of feedback shapes the next recommendation.
Questions in this reading · 12
- What does zero clicks mean for a link that never appeared?
- Can an autoplay sequence identify the objective?
- Who changes the supply of things to recommend?
- Does one more video mean more enjoyment?
- Selection starts before the click
- Which alternatives were available at the time?
- Which side did the person who shared the argument take?
- What does an angry reply signal, and to whom?
- What did the routing system save?
- Low in the ranking or removed from circulation?
- Choosing a nearby lunch and choosing a taste
- Who got what they wanted after three muted seconds?
What does zero clicks mean for a link that never appeared?
Pariser recalls clicking more links from liberal friends and later losing conservative voices from his feed. Imagine two stories shown ten times each yesterday. Today only one gets ten further impressions. How much evidence about preference comes from the other story’s zero clicks today? Click share can move with no change in taste. To interpret it, keep the set of options shown, their audience and their positions alongside the clicks.
Passages and timestamps · 2
I was clicking more on my liberal friends' links than on my conservative friends' links.And without consulting me about it, it had edited them out. They disappeared.Can an autoplay sequence identify the objective?
Tufekci watched a video on vegetarianism and autoplay took her to veganism. She offers retention as an explanation while acknowledging the algorithm is proprietary. Inventory, other viewers or a shared tag could also produce that sequence. A testable question is whether showing different relevant next items to comparable starting histories changes what people later search for voluntarily. Continued playback alone can still be produced by autoplay.
Passages and timestamps · 2
I once watched a video about vegetarianism on YouTube and YouTube recommended and autoplayed a video about being vegan.Now, YouTube's algorithm is proprietary, but here's what I think is going on.Who changes the supply of things to recommend?
Bridle shows children’s titles assembled from surprise eggs, cartoon characters and other popular terms. The phrases read strangely to an adult but make sense to a producer chasing exposure. Now add supply to the earlier loop: exposure changes clicks, and producers respond to exposure by changing tomorrow’s catalogue. To interpret a click-through trend, retain what was eligible each period and compare the choices viewers make when selecting the next item themselves.
Passages and timestamps · 2
So the main way people get views on their videos -- and remember, views mean money -- is that they stuff the titles of these videos with these popular terms.all of these words from other popular videos into your title, until you end up with this kind of meaningless mash of language that doesn't make sense to humans at all.Does one more video mean more enjoyment?
Lembke's example from watching American Idol puts pleasure from the clip next to the comedown after stopping. Huberman likewise emphasizes the baseline preceding an experience. On a recommender's side of the screen, either story can leave the same event: another play. The comparison identifies missing information in a log; it does not diagnose a viewer from clicks. What would you ask that viewer ten minutes after they closed the player?
Passages and timestamps · 3
When I watch YouTube videos of "American Idol," it tips to the side of pleasure.their balance is still tipped to the side of paindepends on your baseline level of dopamine when you arrive there and your previous dopamine peaksSelection starts before the click
Systrom describes a feed already constrained by what people chose to post, with engagement then affecting rank. A similar first step is easy to miss in assistant training. Altman describes a rater choosing between two outputs, and the interview asks who rates them and on what dimensions. Restore the available set before interpreting a preference. An unseen post or an unoffered third answer could not win that comparison. Repeating the same choice does not fill that gap.
Passages and timestamps · 4
rank the things up higher that get the most engagementThey wanna share things that are gonna get likesshow two outputs, ask which one is betterwhich are the humans and what is the processWhich alternatives were available at the time?
Maté asks what relief a behavior offered in the moment. A street interview supplies a concrete setting: the speaker recalls how vulnerable withdrawal would leave him in a park. Brown’s account of perspective taking suggests a practical question here. Ask the person to describe which options felt available then. A count of repeated actions cannot reconstruct those conditions by itself.
Passages and timestamps · 4
what did it do for you? Relief from what?I'd be vulnerableonly a phone call awayStaying out of judgmentWhich side did the person who shared the argument take?
Grey describes a loop in which a visible argument attracts bystanders and grows more visible. A supporter and an opponent can both share it. Recording either action as a like makes the next recommendation test a preference nobody actually expressed. Keep the share and its accompanying comment as distinct observations.
Passages and timestamps · 2
when you tell the joke to another brainThe more visible an argument gets the more bystanders it draws inWhat does an angry reply signal, and to whom?
Stross argues that an ad-supported platform can gain from a brief burst of anger. Bach asks what a public statement does among peers: it can signal membership. The same reply can count as engagement to a platform and as affiliation to an audience. Inferring appetite for more of the same material requires evidence the event log alone does not supply.
Passages and timestamps · 4
common implicit Paperclip Maximizer goal: to generate revenuestructured to reward frequent short bursts of interactionopinions start to signal group loyalty in societies.the beliefs themselfs can change our evaluation of the belief.What did the routing system save?
The suggested route saves the speaker a minute while adding four miles and a toll. Even a perfectly accurate ETA would not establish a traveler’s willingness to pay. Turn the recommendation into a question: for this particular trip, how much money and mileage is a minute worth? Different answers support different choices from the same prediction.
Passages and timestamps · 2
my evidence is not scientifican extra $1 each way all to save me exactly one minuteLow in the ranking or removed from circulation?
In Dorsey’s account of the platform at the time, ranking and account restrictions can both result in someone not seeing a post. First check whether it remains eligible, then its position among candidates. Recomputing a ranking and reviewing an account appeal are also different correction paths.
Passages and timestamps · 2
the algorithms rank and order the conversationThat's why we need an appeals process.Choosing a nearby lunch and choosing a taste
The Japan food video follows someone with little time to buy lunch. Walking distance and what is already on the shelf constrain the menu before any choice. A purchase therefore reflects access as well as taste. To compare preferences across a street, store or lunch break, first record which alternatives were available together. The convenience-store examples alone do not explain population-wide body weights.
Passages and timestamps · 2
these places are in walking distance.just as easy and convenientWho got what they wanted after three muted seconds?
Kurzgesagt describes Facebook’s view rule at the time and copies of its own work. Someone scrolling slowly past muted autoplay could add a count without bringing the creator attribution or income. Chomsky’s model asks what the buyer of advertising purchases: access to an audience. Keep separate entries for exposure, intentional viewing and creator compensation. Logging one does not fill in the other two.
Passages and timestamps · 4
A view is counted after three secondseven if the video is mutedThe market is advertisersthat raises advertising rates.Try it on your own example
With only a click log, how could you distinguish greater preference from greater exposure? Name the denominator you would need, then design a comparison that separates the explanations.