As has been belabored, these AIs are just models, which also means they are only software. Would you be so fire-and-brimstone if startups were using software on healthcare diagnostic data?
> Very few are paying any attention to the 2-10% of safety problems when the AI probability goes off the correct path.
This isn't how it works. It goes on a less common but still correct path.
If anything, I agree with other commenters that model training curation may become necessary to truly make a generalized model that is also ethical but I think the generalized model is kind of like an "everything app" in that it's a jack of all trades, master of none.
> these AIs are just models, which also means they are only software.
Other software are much less of a black box, much more predictable and many of its paths have been tested. This difference is the whole point of all the AI safety concerns!
Clearly the service is not designed for people to only engage with a few folks, it's meant to be a zeitgeist firehose. If you're only following a few people it's like using a spreadsheet for tracking household frozen pizza inventory.
The tool might be more sophisticated than you need but following only a few people is totally fine and should not be overrun with algo content just to promote ad revenue to the platform owner. Maybe the people you want to follow are only on said platform so you are required to consume it that way.
So someone has to say "yes, please commit this change to resolve the issue", that hardly feels like a role remains here, just the DBA manager is needed now.
It's Sapir-Whorf applied to programming: people developed programming constructs that mimic their society structure so while borrow checking may have been developed as a response to problems programmers face directly when developing software, they approached the problem by thinking of their programs in ways that mimic a market-based society and developed solutions that reflect market-based society solutions.
Just like Sapir-Whorf, likely the influence of our society isn't as pronounced as they're suggesting but there's likely a grain of truth that many people approach problems by using ideas from their collected experiences. Similarly, approaching problems in different ways from what we're used to can also be equally challenging due to our lack of prior experience.
They just used a common word that was vaguely applicable to an almost entirely unrelated abstract system of object lifetime verification. Ownership has basically the same semantics as C++'s destructors and moves, just enforced statically. It isn't people looking to the real world for a real-world solution for programming problems.
It is a interesting idea, although there may be some problems too. Maybe it will help, though.
Note that X window system already has the ability for programs to specify a minimim and maximum size hint, aspect ratio, and other things. This isn't new (maybe in Wayland it is new, but in X it isn't new since it is already there).
Honestly I'm not at all convinced, it seems to be another try to push childish stuff in modern systems... Yes, widgets based GUIs are designed with a specific aspect ratio and size in mind, DocUIs tend to be designed with an aspect ratio but scale well enough (think about web-apps, not just Emacs buffers). The lack of integration is only solvable with the classic model "the OS is an application, there is a single unique environment for anything", so all apps are written as part of the sole system/framework from the most lower level to the highest end-user programming.
Something perfectly possible, actually done in the past, with very nice results, see just for instance https://youtu.be/RQKlgza_HgE to not go too much back in history toward Xerox Smalltalk workstations. but also something commercial big player do dislike, since in this model you can only pay the programmer, not the program/service so no giants only competition...
Text and relative text rendering is the most flexible and integrable tool we have so far, and the web itself is a good proof and show as well the limits of trying not giving power to the end user.
On "modern desktops", honestly Ubuntu Unity was a thing because the side bar was just a quick launcher with some icons, something usual on a good place since we have larger and larger monitor with less and less vertical space, and the top bar was as small as fluxbox at al. Gnome SHell tried the narcissistic way, forcing people to see animations just for common windows switching witch surely have a kind of WOW effects for the kids, but far less interesting to work all day. Ubuntu also pushed the HUD, or a way to replace the apps menu with a CLI search&narrow style, witch was essentially the same principle than dash and quicklaunchers in general AND the Emacs M-x model. However in Emacs anyone can craft a function in a snap and run it through M-x, in modern desktops it's a hard, long thing with much boilerplate. Just to create a .desktop, witch is a kind of functions skeleton in the Emacs model, you need a dedicated file only for that, put in a specific place, than the code somewhere else etc. People do like seen beautiful gardens, but do want quick things, so the path of regulate the world in advance all modern system try is a failure in advance.
> The lack of integration is only solvable with the classic model "the OS is an application, there is a single unique environment for anything", so all apps are written as part of the sole system/framework from the most lower level to the highest end-user programming.
This is one reason why I wanted to design an operating system, although I also wanted to change many things from what other systems do, although there are also some similarities to features of other systems.
My ideas do involve better integrations between parts of the system (although different parts can still be individually reimplemented and replaced), and my ideas involve both low-level and high-level programming.
> Gnome SHell tried the narcissistic way, forcing people to see animations just for common windows switching witch surely have a kind of WOW effects for the kids, but far less interesting to work all day.
Sometimes animations may be helpful if objects are moving around on the screen and you want to easily see where they are moving to, but a lot of the animations are worthless and I would rather not see them. If an option is added to allow disabling all animations (and to adjust their speed), then that will be a good idea.
> However in Emacs anyone can craft a function in a snap and run it through M-x, in modern desktops it's a hard, long thing with much boilerplate.
It does look like a good idea, if anyone can craft a function and run it (which is also an idea of what I might intend in my operating system design). I don't use a desktop environment on my computer and do not use .desktop, although some programs do.
There's a collection of techniques in the mental health field called Cognitive Behavioral Therapy that is about helping people identify and replace these negative stories. If folks are having a hard time applying these ideas themselves alone, look into this therapy as it can be good to have a therapist help you through using the technique.
My main fear with AI becoming our go-to information retrieval platform is how can we be sure that the answer is objective truth without any slant? Shouldn't there be an ethical obligation that a model should not only fact check but also provide alternative theories? We lost that somewhere in journalism but we could regain that if we prioritized that the thinking and deciding still needs to occur human-side.
Go try to ask any of these hosted LLM bots about sexual topics to find out how far from objectivity we are. I earned a stern warning after trying to write a dirty pun with ChatGPT
Imagine trying to do research on sexually transmitted diseases, or trying to self-educate about sex. Unfortunately the limits put in place are quite coarse today.
> Very few are paying any attention to the 2-10% of safety problems when the AI probability goes off the correct path.
This isn't how it works. It goes on a less common but still correct path.
If anything, I agree with other commenters that model training curation may become necessary to truly make a generalized model that is also ethical but I think the generalized model is kind of like an "everything app" in that it's a jack of all trades, master of none.