That matches my experience with non-coding tasks: it’s not very creative, but it’s a comprehensive critical reader.
5,070 karma · joined September 25, 2009
It’s an indictment of how decisions are made at those companies, not a vote for the relevance of chatbots.
They could argue that the decision was made based on declarations that did not align with the private conversation that Zuckerberg had at the time, as those emails came out since.
If you give me extra wishes, I’d love three options, to either say that the ad is annoying, the brand isn’t for me, or I don’t want to be offered that type of product.
The quality of ads would skyrocket if I could just stop seeing efforts to get me to be interested in things that I will never buy.
Just before joining Facebook, I was living abroad and confronted to ads in a language I didn’t understand constantly. As my bootcamp task, I measured that this was 4% of ads shown to users. At the time, this was already billions of dollars. My manager deemed that to be a ridiculous and pointless exercise. One night at the bar (there were three bars in the London office at the time), I mentioned it to a guy who happened to be the big ad boss, who immediately prioritized the project, I got a couple of smart guys who joined after I was promoted for finding this.
A bit later, I checked the conversion rate by how many times you’ve seen the same ad before. It was a precipitous cliff: people click on things they’ve never seen before. Ranking ads from the same advertiser happens to be one of those SQL/Hive query that doesn’t scale well, so I had to use the fact I came to the office early and has 12 hours of uninterrupted server time before the daily queries were hammering anything, and I had to sample a lot—but I realized I could sample by server, which helped a lot.
I tried to mention it to the same guy, who said he knew about that problem, but empowering users like I suggested would not work: it would shrink the matching opportunities, AI was getting smarter, etc. In practice, the debate around privacy got very toxic, and Facebook couldn’t let people do that without some drama about storing a list of advertisers that they said they didn’t like.
One of Sandberg’s trusted lieutenants lost a child late in her pregnancy; it was a whole thing. She started seeing ads for baby clothes just after, which triggered an optional ban on alcohol, gambling, and baby stuff. That’s still there. I worked with her briefly a bit later (after months of bereavement) and asked if it made sense to expand the category. She replied that those were two legal obligations, plus her well-known personal drama that no one dared push against, so she was able to push for those three, but that the company had changed. No other categories could be added: at that point, it would be too difficult. Mark used to not care about ads, but he started having expensive ideas, notably AI (to ban horrendous content); he needed the money, and he started to care about raking as much dough in as possible. I had worked on horrendous content (instead of ad language) enough to know that it mattered, so I was very conflicted. It felt surreal how much things had changed in nine months.
All that still feels like a giant waste, not the least how much energy goes into making and showing ads to people repeatedly swearing at their screen, begging to make that annoying copy disappear.
If I were to hide something fishy, I’d define the shared coverage to include zip codes with fewer human accidents:
* zip codes are relatively small, so you have a lot of variance per code due to randomness of rare events;
* Waymo mostly covers urban areas and nearby suburban areas, so you can claim suburban zones were only recently covered by Waymo or that the insurance had too few members in certain zip codes to include them.
I don’t think that study is wrong, but it is in Waymo’s interest to claim they have few accidents and in insurers’ interest to claim there have many, or rather more, accidents to raise their premiums.
Churchill is similar and possibly more familiar to the English-speaking public: presented as a lion against Hitler, but don’t dig too much towards what he did with Irish republicans or around India’s independence.
I don’t think FDR was that much different if you read what he privately wrote about Jewish immigrants to the US in the 1940s or the internment camps of Japanese-Americans.
All three are the modern keystones of human-rights-defending countries, but the second best-known thing about them is how much effort they put into planning openly racist policies, up to genocides. You can read Hannah Arendt and re-read after accounts on what happened during the Battle of Algiers, the Bengal Famine, or Manzanar, and it hits differently the second time.
If Charles taught anything to anyone listening, his children, troops, voters, or readers, it would be precisely that.
The fact that you don’t know and people offer dozens of credible guesses should probably tell you the problem is widespread.
Soon after Musk took over, I started having accounts wishing me violent deaths, repeatedly commenting on everything I said with graphic details (broken bones, poisoning, dragging my body over the pavement, etc.). That happened occasionally before, but they typically got banned. After the takeover, those were gone (and my account got blocked a couple of times for quoting them).
That hasn’t happened after a year on BlueSky and Threads.
Scams were rampant on large accounts and people looking at cryptocurrencies: more than three-quarters of comments were obvious patterns that I had flagged dozens of times. I noticed those earlier today on Threads; let’s see if they reappear and make up most of the discourse there.
Another thing: Eppo is actually more focused on Experiment (A/B testing) analysis, and we let customers use any feature flag they want. Some clients use an in-house solution; some use Eppo, and many use commercial third-party options. If those systems have issues, our monitoring is often the first to flag it, and those errors are odd, so our clients pretty systematically flag us when that happens. That long explanation to say: I’ve seen a lot of how feature flagging system fail. Several a day, every day, for months. That’s why I can confidently say that Eppo is great, LaunchDarkly is great. Other commercial solutions… I have doubts: they work for basic use cases, but as soon as you have ad blockers, a bad connection, multiple ways to identify users (cookies vs. accounts; same account on mobile vs. desktop browser), or users who know how to edit their cookies, things get bad. In-house solutions (with one exception) also have common oversights.
We talk a lot to experienced engineers who think they can build it —and they can, key patterns are simple— but tricky cases always pop up. I usually ask a few questions to clarify and help them understand what those gotchas are. Senior tech leads aren’t always… keen on thinking, “That’s too hard for me.” So if they push back even lightly, I tend to recommend people to go that way and expect them to come back weeks later, asking tough questions. It’s never lost time: they learn a lot about their own architecture, limits of spinning up new services, passing configuration, design, etc. More importantly, they learned that third parties have relevant experience building this.
Because lawyers are in the business of managing risk, and knowing what OC was unhappy about was very much relevant to knowing if he presented a risk.
There was one thing that I cared about (anti-competitive behavior, things could technically be illegal, but what counts is policy so it really depends on what the local authority wants to enforce), so I asked a lawyer, and they said: No way this agreement prevents you from answering that kind of questioning.
I don’t doubt that it’s true in OC’s case. But I don’t think it’s representative of the impact of technology. It’s the example that will get Bill Gates’ attention and end up on the stage at TED, but it might not be representative of 90% of how people are affected by technology. It might, and the story of how videos will change how people relate to each other (what made Chris Anderson buy TED, what Salman Khan has built, what Grant Sanderson a.k.a, 3blue1brown has done). It is true, anecdotally. It is a fantastic, compelling, clear story. I love that story: video is the easiest way to share any practice; people watch, learn, practice, and get good at something no one in their village knew was possible. Hurray.
But when I trained as a statistician, i.e., epistemologically someone working for the state, I was told I was the opposite of a journalist: my job is not to tell stories that explain what a problem is. My job is to tell the boss how common that story really is. A story tells all the details to pull your emotions, and journalists (at the time, at least) were telling fact-check, double-verified stories. “Truth” wasn’t our goal. It is too easy to get numbers that argue for any position. Representative was what we were told we should be aiming for—so that the government makes the right decision.
I’m not seeing a lot of “representative” here.