184 karma · joined July 18, 2023
My guess: everything that slows down ai development is good, because it gives society time to adapt.
(I think this plan is flawed, because it is easier to adapt to open research than to closed research)
Can confirm. Last year I implemented a simple postgREST server with rowlevel security. (The postgREST logs are really good. With cookbook and all)
The path there was somewhat difficult, but once it worked, it was truly magical. And the mechanisms involved were quite simple even.
1. grab the minified js from the CDM, i.e. https://cdn.jsdelivr.net/npm/canvas-confetti@1.9.2/dist/conf...
2. remove the comments at the start and end
3. paste it into the template below
javascript: (async () => { /* paste minified JS here */ ; confetti();})();
4. (optional) customize the confetti()-call5. copy the result and paste it into bookmark
This works on most pages (tested on some tabs I have open), but it does not seem to work on hackernews itself. Here it gives me ContentSecurityPolicy-Error when creating the worker. (at least it does in edge)
My other guess is, that it really is more expensive. I think, the publisher's digital infrastructure is really bad. And there is a lot of overhead when publishing a new book with lots of DRM on it. And most books don't get sold that often. So a small number digital purchases must pay for that over-proportional up-front cost.
I don't think big companies do such "deliberately" (in the sense that a human made the explicit decision to do it). But when there is a certain behavior that performs better in practice, then such behavior will proliferate.
I say, that when the effect is known, and the company chooses not to reduce it, they are acting deliberately. The purpose of a system is what it does.
I agree, that the relation between the effect and the specific behavior may not be obvious, as you pointed out ("Managing bias is a wildly difficult..."). But it seems that google does not even acknowledge, that this effect exists [1].
Beside the intention, I think the more interesting questions would be:
Do the induce such a bias (deliberately or not)? Do the induce any bias at all?
[1] https://www.politico.com/magazine/story/2015/08/google-2016-...
Is there a quick ways to integrate this to existing sites? maybe some js I can include in index.html for websites I can control? maybe some bookmarklet to inject it into websites I don't control?
I imagine this would also increase the complexity of the python intepreter by more than 1%, which in turn would increase the number of bugs in the interpreter by more than 1%. Which would burn both manpower and CPU-Cycles on a global scale.
(I assume that optimization, that reduce complexity are already exhausted, e.g. stuff like "removing redundant function calls". )
That may become expensive fast
that doesn't reflect my experience at all.
from the outside many weird things seem to have some deeper meaning and thought-out plan. But when you get an inside-perspective, you notice that often no meaning or plan exists, and the reason for the weird thing is something silly like "an important person made a suggestion, that someone else misunderstood, but also made it a priority", or "there was a good suggestion, but it wasn't implemented, because the person who suggested it, was in bad standing with the management".
the implication sounds good, but is easily rejected with a counter example: a good student can surpass his master. A mediorcre student with multiple masters can surpass all of them (tangent: this is why the medival master-journeyman system was so efficient, I think).
Or to make the argument more abstract: Your implication seems to assume, that transferlearning does not exist.
Bonus 1: Since you pay per token, it's much cheaper than a chatgpt abo
Bonus 2: You can increase the context window dramatically (iirc 8000 being the max for playground, while 2000 is the max for chatgpt)
Context: I am interested in writing GPU-code as well (for ML too, but also for custom physic simulation). I also think, that compute-shaders are the way to go. But my intution says, that openCL would be simpler than vulcan.
Question: Have you considered alternatives to vulcan's compute shaders? (e.g. opencl instead of vulcan, or maybe some other ways entirely to run GPU code, such as rust-gpu)
It basically goes like this:
1. Teachers are authorities and present facts to children
2. people later learn that some of these facts are in-fact not factual
3. people learn that:
a) authorities can not be trusted
b) it is ok to make up facts, as long as they sound good
4. people apply a) and b) to other situations> The electron cloud model says that we cannot know exactly where an electron is at any given time, but the electrons are more likely to be in specific areas.
https://en.wikipedia.org/wiki/Artificial_general_intelligenc...
why is that?
I am asking, because so far I have seen old linux users refusing to move because they require some specific x11 feature that is not (yet) supported on wayland. I thought that new linux user wouldn't have such requirements.
could you elaborate?
The main problem I see here, is that management may use ai as an opportunity to push for more micro management. And if that happens, the consequences will be similar to regular non-ai micromanagement (e.g. higher perceived productivity, less actual productivity, developers leaving).
I think having a good architecture plays a big role here.
I guess you gain some security, when you don't have to worry about some things. But you also lose some security, because of the added complexity.
In this scenario unit tests would not add value over integration tests when looking for the existence of errors.
But: In a bigger project you don't only want to know "if" there is a problem, but also "where". And this is where the value of unit tests comes in. Also you can map requirements to unit tests, which also has some value (in some projects at least)
edit: now that I think about it, you can also map requirements to e2e tests. That would probably even work much better than mapping them to unit-tests would.