1,219 karma · joined March 3, 2015
I was also not aware GitHub let you run actions on a schedule, that could also work - but I guess you have to use artifacts for the persistence?
> But with the advent of AI in almost all tools these days, I'm wondering if its more accessible now and building/publishing a game is easier than ever ?
Also worth noting that storefronts like Steam and Itch expect you (but obviously can't force you) to disclose whether you've used generative models to create assets - which turns away people who consider them unethical or a mark of shovelware
Linked in the article, about her meeting with Hunter
https://genderanalysis.net/2023/11/new-trial-exhibits-in-doe...
> This is just straightforward attempts at guilt by association.
Call him a "government staffer" all you like, let's call it what it is: her meeting with an explicitly religious interest group
> A followup email from Hunter indicates that he met with Cass on September 22, 2022 (ECF 184-1). Even as Paul Vazquez wanted to invite Cass to the Board of Medicine hearing as a subject matter expert, Hunter felt that would put her “in a difficult position”:
Why downplay this link in a discussion about the politicization of science?
Cass meeting with DeSantis staffers [1], for one
[1] https://www.erininthemorning.com/p/cass-met-with-desantis-pi...
> Notably, Hilary Cass met with Patrick Hunter, a member of the anti-trans Catholic Medical Association who played a significant role in the development of the Florida Review and Standards of Care under Republican Governor Ron DeSantis.
They note that existing age verification setups largely either rely on providing ID, or on a combination of manual and automated behavior profiling (face recognition, text classification, reports from other users), both of which have obvious privacy and/or accuracy issues. The "double-blind tokens" point to a summary by LINC explaining how they _could_ be implemented with zero-knowledge proofs, but I could not find an article or a practical implementation (could just be a mistake on my part, admittedly)
At _best_ you end up with a solution in the vein of Privacy Pass - https://petsymposium.org/popets/2018/popets-2018-0026.pdf - but that requires a browser extension, a functioning digital ID solution you can build on top of, and buy-in from the websites. Personally, I also suspect the strongest sign a company is going to screw up the cryptographic side of it is if they agree to implement it...
Making training exactly reproducible locks off a lot of optimizations, you are practically not going to get bit-for-bit reproducibility for nontrivial models
You joke, but I worry about this if Norway implements an age limit based on geographic location - it either gets kids onto the shittiest possible free VPN which MITMs everything, or gets them onto imageboards/apps with even fewer facilities for handling abuse than FB and Snapchat have
E: > Also wouldn't HN count as social media too, as we are interacting here?
Family minister Toppe was immediately asked afterwards if the age limit would cover newspapers with comment sections, and she responded that it would - so parents would have to consent to you sharing personal info with the newspaper, which seems like a long way to say "your parents decide whether you get to read the news"
https://www.vg.no/nyheter/i/mPzg0q/regjeringen-vil-innfoere-...
Again people conflate correlation and causation very easily here, and don't account for changes in how mental health conditions are coded, see https://www.thedailybeast.com/the-coddling-of-the-american-p...
> no plausible alternative culprit offered [for plummeting birth rates]
"More than half the drop in America’s total fertility rate is explained by women under the age of 19 now having next to no children. Around a third of the missing births would have been unplanned, and the majority of them would have been to women on low incomes."
The activist movement championing the age limit are explicitly framing the issue (as Haidt does) as social media use directly leading to worse mental health markers in children and teens, and arguing that this harm is so great that any proposal claiming to fix it should be implemented, consequences be damned - exactly the kind of "think of the children" argument that gets people to argue against end-to-end encryption.
> We have very little understanding of the impact of an always on skinner box/rage box, and I think it's fair to want to understand more about how that could impact developing minds and to be more cautious about how we deploy systems that could be significant harms. > (though I'd frame it more around regulating what interactions social media can offer.)
I argue that the problem is that advertisement and VC funding are fundamentally antithetical to making good spaces for discussion. Making a good space requires moderators who are invested in participating themselves while being conscious of their role - the kind of work where the workload scales linearly with users, but the expected accuracy and severity of screwing up grows more rapidly. However, your investors want more users and eyeballs on ads as fast as possible, while making moderation work as cheap as possible.
Social sites don't have to be "always-on Skinner boxes", but it comes down to technical and organizational decisions and whether you are willing to treat discussion spaces as a common good which don't need to run with billions in profits
It's very telling that Toppe treats privacy concerns as just something for the nerds to handle ("et lenger lerret å bleke" [2]) rather than a fundamental risk introduced by any attempt to implement this in a centralized fashion - with the Norwegian government already relegating important authentication to banks [3], I remain confident that any attempt to implement this will be 1. trivially circumventable 2. annoying as shit to the average user 3. an increased risk to kids trying to escape abusive situations 4. a farce akin to the UK porn passes [4]
I wish a fraction of this effort could have gone into maintaining good online spaces which don't relegate moderation to the first BERT classifier available off HuggingFace or remote workers being paid next to nothing. Aftenposten getting their FB posts deleted in 2016 [5] could have been an excellent time to start a discussion about not hitching Norway's digital infrastructure to ad companies, but they didn't - and now we are left with people assuming that social media necessarily has to handle bullying and advertising in a shitty way, because the big sites deliberately design towards the lightest possible automated/piecemeal moderation for hundreds of millions of users.
[1] e.g. "my depressed kid is on the phone all day, because The Algorithms(TM) are beaming depression into their head, rather than providing an outlet for them"
[2] https://www.vg.no/nyheter/i/mPzg0q/regjeringen-vil-innfoere-...
[3] Yes, there is MinID, but many services require BankID, turning the first few weeks for people immigrating - _before_ they have a Norwegian bank account - into an absolute barrel of laughs
[4] https://www.independent.co.uk/news/uk/home-news/porn-passes-...
[5] https://www.nrk.no/kultur/aftenposten-sin-facebook-kritikk-f...
Skimming the article I wonder how extensible the approach would be to other context-sensitive translations - the silly example that came to mind is whether to use tu or vous and the corresponding verb conjugations in French - as the algorithm described isn't limited to binary categories
[1] https://www.thedrive.com/tech/34817/this-25000-game-boy-is-m...
[2] The aerial is kind of a giveaway, admittedly.
If you want a MIDI pad controller, the Korg nanoPAD 2 [1] is $65 new (and often pops up used) and powered entirely off USB. You can then set up Sitala [2] to listen for notes from the pad and drag-and-drop samples onto each note.
If you want keyboard-style keys, the Pimoroni Keybow kit [2] is $57 - there is a bit of assembly, but no soldering.
[1] https://www.korg.com/us/products/computergear/nanopad2/
One recent attempt at it is Mesh Anything [1], which tries to generate triangle meshes with sensible topology[2] for a given point cloud - but it's still in the early stages, it can fail somewhat dramatically on meshes with smooth and concave parts [3] and it has a hard cap on the number of triangles it can generate.
[1] https://buaacyw.github.io/mesh-anything/ (submitted previously as https://news.ycombinator.com/item?id=40701992 )
[2] although they use the somewhat awkward term "artist-created mesh"
[3] e.g. https://i.imgur.com/z52QeiQ.png
ONNX [1] is more or less this, but the challenge you immediately run into is models with custom layers/operators with their own inference logic - you either have to implement those operators in terms of the supported ops (not necessarily practical or viable) or provide the implementation of the operator to the runtime, putting you back at square one.
[1] https://onnx.ai/