James Somers is awesome. Every now and then, I see another of his articles that comes up here. Really great writer, who didn't set out to be one, but switched from a career in tech I believe.
This is super interesting. I've been visiting HN every day, multiple times a day for the past year. Over the last month, as I was preparing for interviews I removed the HN bookmark on my browser tab, and almost for weeks in between interviews I would go without checking it.
And honestly I think I was healthier, and happier. Reading, and bookmarking things on HN overflowed my brain with information, and created more anxiety in me I think. Just visiting back to see what I missed after not checking it the past week, and I was better off without it I feel.
Kinda sad that Netflix opted for quantity over quality. It shows with all the inane sequels like Murder mystery 2, Extraction 2 and with the idea of just filling an uninteresting script with some stars to make the numbers like Red Notice. And many of the top 50 shows reflect just that. The last Netflix original I watched and liked was The Queen's Gambit.
You should try "Obliterated", its hilarious, it doesn't take itself aerious, it plays with almost all TV tropes we know (in a good way), references everything from Rambo to Die Hard and has an eye for detail (every scene hints at something or is important in some regard). Also, episodes 2-8, each roughly an hour, cover an in-series time frame of 7 hours, which is a bice touch. Throw in good acting, good dialogs and liekable characters and you uave a great show.
> but in reality they just couldn’t handle the data
If you mean from a "critique" standpoint then no. If you meant from a "data wrangling standpoint then yes, as binary like/dislike data is just a lot easier to work with for recommendation algorithms.
There are many good reasons to move from a five star rating system to a binary rating system, and yeah, I think many of those fall into the "people lied to themselves" category:
- People tend to note use five or ten star systems on a continuous scale with certain points on the scale being biased
- People tend to go into "movie critic"-mode when they see a 5/10-star scale, as those are usually used on sites like IMDB. That drives them to try and rate the movie "objectively" and in accordance with an intellectual image they want to portray, rather than what they actually like/dislike consuming and spend their time watching
- Netflix also displayed the ratings as 5-star "adjusted ratings for the viewer", which already took your preferences into account. Not a single person I've talked to back then was aware of that, so everyone tried to do the same mental gymnastics they do when trying to project global IMDB ratings to their personal preferences. Moving to a "XX% match for you" together with the like/dislikes is something that people understand a lot better
All-in-all, I don't think the rating system really has been an issue in the recent years. The catalog has been a much bigger issue during that time. I'm pretty sure that Netflix's rating and recommendation system has been good enough that it has served me everything that I'd like to watch on their platform and now I'm out of content.
> So to summarize, customers are dumb & recommendation algorithms are hard.
Or to not throw away all the points I argued for: There were a lot of good reasons for moving away from the five star systems, and the main motivation was providing a better less confusing UX
> I loved Netflix stuff when they started out, banger after banger was delivered. Something changed.
The main thing that changed was that media companies woke up to streaming and stopped giving away the rights to properties that many people enjoy (mainly nostalgia shows + blockbusters) for cheap. That forced Netflix much more into media production of their own, which has it's up- and downsides, and gave Netflix the same warts that media companies always had (e.g. having to make hard decisions around canceling shows). It's a very clear case of a first-mover losing its advantage over time.
Netflix, famously, had a public contents to try to improve the recommendation system, with very little success. [1] That system was a demonstrably good system, that maximizes preference, but only if you rate a bunch of content. I'm not aware if it's still in use.
But, I've rated many hundreds of shows/movies, and the rating is very accurate, for me. Very biased and scaled, but completely predictable/reliable.
In my case, the rating can be corrected with Netflix 7 being my 0, and Netflix 10 being my 10, mostly linear. For a Netflix show, 8 is my 0, since those are biased a bit more.
Although, technically, the most profitable customer is the one that has low view time, but still subscribes. The most stable customer is probably one with high view time, but they're also probably the least profitable, unless they're naturally helping promote the service.
But they are not giving people what they want. Netflix recommendations are notoriously bad and Netflix is not all that successful lately with its own shows. Even when they have actually good shows you would like, you rarely find them through Netflix recommendations - you need someone to tell you about the show.
Wealth begets wealth. Capital compounds with no effort. A reminder for myself to revisit Capital by Thomas Piketty from 2013(!)
"When inequality gets too extreme, then it becomes useless for growth, and it can even become bad because it tends to lead to high perpetuation of inequality over time and low mobility."
Funnily enough, I was watching the Spotify series on Netflix, The Playlist, and saw Sony Music Group's fight with the Pirate Bay(which covers similar themes) before Spotify exploded on to the scene. This was a real event too covered in the early episodes - https://en.wikipedia.org/wiki/The_Pirate_Bay_trial
The board was a non profit board serving the mission. Mission was foremost. Employees are not. One of the comments a member made was, if the company was destroyed, it would still be consistent with serving the mission. Which is right.
The fallout showed non-profit missions can't co-exist with for-profit incentives. And the power that investors were exerting, and employees (who would also benefit from the recent 70B round they were going to have) was too much.
And any disclaimer the investors got when investing in OpenAI was meaningless. It reportedly stated they would be wise in viewing their investment as charity, and they can potentially lose everything. And there was an AGI clause that said it will reconsider all financial arrangements, that Microsoft and other investors had when investing in the company was all worthless. Link to Wired article with interesting details -https://www.wired.com/story/what-openai-really-wants/
> The board was a non profit board serving the mission. Mission was foremost. Employees are not.
They need employees to advance their stated mission.
> One of the comments a member made was, if the company was destroyed, it would still be consistent with serving the mission. Which is right.
I mean, that's a nice sound bite and everything, but the only scenario where blowing up the company seems to be consistent with their mission is the scenario where Open AI itself achieves a breakthrough in AGI and where the board thinks that system cannot be made safe. Otherwise, to be relevant in guiding research towards AGI, they need to stay a going concern, and that means not running off 90% of the employee base.
> Otherwise, to be relevant in guiding research towards AGI, they need to stay a going concern, and that means not running off 90% of the employee base.
That's why they presumably agreed to find a solution. But at the same time shows that in essence, entities with for-profit incentives find a way to get what they want. There certainly needs to be more thought and discussion about governance, and how we collectively as a species or each company individually governs AI.
I don't really think we need more thought and discussion on creative structures for "governance" of this technology. We already have governance; we call them governments, and we elect a bunch of representatives to run them, we don't rely on a few people on a self-appointed non profit board.
> One of the comments a member made was, if the company was destroyed, it would still be consistent with serving the mission. Which is right.
I know you're quoting the (now-gone) board member, but this is a ridiculous take. By this standard, Google should have dissolved in 2000 ("Congrats everyone, we didn't be evil!"). Doctors would go away too ("Primum non nocere -- you're all dismissed!").
Indeed, it made no sense. But that's why I never attach any value to mission statements or principles of large entities: they are there as window dressing and preemptive whitewash. They never ever survive their first real test.
Yep, this is spot on. The entire concept of a mission driven non profit with a for profit subsidiary just wasn't workable. It was a nice idea, a nice try, but an utter failure.
The silver lining is that this should clear the path to proper regulation, as it's now clear that this self-regulation approach was given a go, and just didn't work.
It's not necessarily consultants in these companies that are malicious. I find working with many to be honest, and deeply caring about their work. It's that some of the management in these companies are incentivized to engage in duplicitous behavior to gain clients one way or another.
Working for a tech consultancy before, they do have clear lines to distinguish for legal reasons and make you take courses. For e.g., you can't give or receive gifts of any monetary value, etc. But management find other ways, to gain client's trust and affection. One way is to overwork the consultants, and other is to share information that is privy sometimes.
