From 3/24/2012, 8:15:00 PM till now, @minimaxir has achieved 51729 Karma Points with the contribution count of 7542.
Recent @minimaxir Activity
The entire existence of the Facebook Transparency report is as a counter to the Facebook Top 10 Twitter account (https://twitter.com/FacebooksTop10) showing that Facebook tends to surface right-wing content.
Facebook prefers to be known for facilitating spam than right-wing content.
Years ago, I YOLO quit Apple after my professional development stalled (HN discussion then: https://news.ycombinator.com/item?id=14270897) and I felt I had enough in my personal data science portfolio that I could leverage it to get a job in the industry. I had overestimated my hireability but I eventually did get a DS job after four months, and am still at that company.
In retrospect, it was correct from a career standpoint, but I will not quit a job again without having another job lined up, especially since for data science in particular the industry is competitive and would result in another prolonged job search.
Yes, but those are open to very broad and potentially inconsistent interpretations.
AI text content generation is indeed a legit industry that's still in its nascent stages. It's why I myself have spent a lot of time working with it, and working on tools for fully custom text generation models (https://github.com/minimaxir/aitextgen).
However, there are tradeoffs currently. In the case of GPT-3, it's cost and risk of brushing against the Content Guidelines.
There's also the surprisingly underdiscussed risk of copyright of generated content. OpenAI won't enforce their own copyright, but it's possible for GPT-3 to output existing content verbatim which is a massive legal liability. (it's half the reason I'm researching custom models fully trained with copyright-safe content)
It's very good that OpenAI is relenting and opening up the API; however the Content Guidelines are still too onerous such that even if you can think of a good use case it will be a liability at best even if your app gets approval.
At this point (1.5 years later), if you're looking to make a sustainable business on AI text generation, you may want to experiment working with large-but-not-as-large models like GPT-J-6B; it'll be much cheaper too in the long run.
This is creepy in the best case (which would cause me to ignore the recruiting email) and ethically dubious in the worst case (which would also cause me to ignore the recruiting email, but also call out the company on social media for doing so).
This blog post mostly makes me think less of both the startup and the investors. Deepfakes are a delicate issue and this post treats them more as a fun opportunity.
Context: https://twitter.com/OBSProject/status/1460782968633499651
> Near the launch of SLOBS, @streamlabs reached out to us about using the OBS name. We kindly asked them not to. They did so anyway and followed up by filing a trademark
> We’ve tried to sort this out in private and they have been uncooperative at every turn
Streamlabs will be removing the OBS name from its Streamlabs OBS product
10 points • 1 comments
Netflix elaborates on this in their blog post: https://about.netflix.com/en/news/new-top-10-on-netflix
> Figuring out how best to measure success in streaming is hard, and there’s no one perfect metric. Traditional measures like box office or share of audience (which was designed to help advertisers understand success on linear TV) aren’t relevant to most streamers, including Netflix. Having looked at the different options, we believe engagement as measured by hours viewed is a strong indicator of a title’s popularity, as well as overall member satisfaction, which is important for retention in subscription services. In addition, hours viewed mirrors the way third parties measure popularity, encompasses rewatch (a strong sign of member joy) and can be consistently measured across different companies.
Years ago, Netflix got a lot of flak for using "users who saw atleast 5 minutes of a series" as an indication people watched it. Quantifying video performance based on watch time has become more industry standard (especially on social media which loves autoplaying videos) as it accounts for retention.
Arcane is an odd case as it’s released in batches of 3 episodes instead of being full-dropped as most other Netflix originals.
It's fun to look at the numbers for Squid Game (under TV/Non-English) over time. The first week on September 13 was better than most English TV shows...and then the next few weeks nearly 10xed that first week performance. Even now, two months later, it beats most of the Top English TV shows.
And reach out to 460 people for professional networking. And 230 public commits on GitHub. And various unspecified other criteria (which are present in the PDF linked by sandofsky, including always being available for interviews and job hunt audits).
Netflix releases weekly lists of most popular titles by watchtime
5 points • 0 comments
Lambda School is renaming itself to Bloom Institute of Technology
45 points • 37 comments
More generic / not unique content, in a domain commonly associated with blogspam.
The first thing Meta should accept about VR is that people are going to have extremely kinky sex in it and there's nothing they can do about it.
Making the dislike count private across YouTube
1439 points • 1468 comments
Pages/Numbers/Keynote do support versioning in-app when the doc is stored in iCloud Drive, albeit not as efficient as true versioning.
Per Down Detector, it is mostly Xfinity: https://downdetector.com/status/xfinity/
Cell networks in San Francisco are overloading.
EDIT: Starting to come back now.
Widespread internet outages in the entire SF Bay Area
24 points • 7 comments
site design / logo © 2021 Box Piper