dfbrown

One of my big questions since the HuggingFace incident is whether OpenAI would have been so forthcoming (the independent METR report, the Black Hat talk, OpenAI’s own report) if the incident had been discovered internally and not publicly.

Not disclosing this despite apparently knowing for weeks makes me think they would not have, or would have concealed details, or delayed disclosure. Combine that with their technical missteps that led to this (weak sandboxes, very slow to detect the misbehavior) and I now strongly doubt OpenAI is capable of responsibly developing such potentially dangerous AI systems.

Their collaborator's report includes a more significant issue, an RCE on a mintlify server: https://kibty.town/blog/mintlify/

How real is it though? This blog post says

In this post, we’ll explore some of the prompting approaches we used in our Hands on with Gemini demo video.

which makes it sound like they used text + image prompts and then acted them out in the video, as opposed to Gemini interpreting the video directly.

https://developers.googleblog.com/2023/12/how-its-made-gemin...

"The orders instruct him to not make public any information that would identify A.B., or the medical professionals involved, to call A.B. by the child’s preferred name and gender pronoun, and to not share his opinions of the case publicly."

"In June 2020, C.D. gave an interview to a YouTube channel, where he’s alleged to have identified health-care providers, revealed information about A.B.’s mental health, medical status or treatments, and gave out information that could reveal C.D., A.B. and the mother’s identity."

"“This order should not restrict C.D.’s right to express his opinion in his private communications with family, close friends and close advisors, provided none of these individuals is part of or connected with the media or any public forum, and provided C.D. obtain assurances from those with whom he shares information or views that they will not share that information with others,” the court said."

It sounds like his arrest is more due to him repeatedly discussing the case in public rather than the pronouns he used.

https://nationalpost.com/news/b-c-father-arrested-held-in-ja...

There are two separate things here, interpenetration due how contacts/constraints are solved, and whether you do continuous collision sub-stepping or not.

For stability and performance reasons physics engines usually have parameters that soften or add compliance to contacts/constraints. A bit of compliance is almost always better than infinitely stiff collisions/constraints. There are cases where infinitely stiff systems either have no solution, are very expensive to solve, or would produce very extreme impulses (causing things to explode). Including some compliance fixes these issues. It is also often required to produces more realistic looking results since objects in real life aren't infinitely stiff, they either flex or break.

For performance reasons most physics engines also do not completely solve their constraints. They either use a fixed number of iterations (most common, including the demos here) or solve up to some specified error threshold. This tends to add some additional compliance to complex scenes (stacks/piles of objects for example).

With the right parameters a good rigid body physics engine should be able to prevent noticeable interpenetration in most situations, though the performance cost may not be worth it. In these demos if you max out Position iterations, velocity iterations, and frequency you should see significantly less interpenetration.

As for continuous collision detection/sub-stepping, this is a very common feature to prevent fast moving objects from clipping into or tunneling through other objects. However resting/continuous contact cannot be handled by stepping to the next intersection time and so have to be handled differently. Also multiple simultaneous or near simultaneous collisions can grind things to a halt in degenerate cases (such as multiple stacked objects that are almost, but not quite in resting contact). This is why physics engines that support continuous collision usually let you set a maximum number of sub-steps.

All but two state's electoral votes are decided by the result of the popular vote in that state, congressional districts have no effect on that.

Local and proportional representation is the main reason for having the house of representatives, while the reason for Gerrymandering is to disproportionately favor one party or another.

No, ranges in C++ are analogous to the iterator trait in rust

I'm not an expert in the topic, but my understanding is RGB is a poor color space for computing color difference. This could be why your mosaics end up so washed out. [1] suggests using a CIELAB color space [2].

Edit: Looking at the code more closely it looks like you were using Lab at one point but commented it out[3], so I'm guessing you're already aware of this.

1: https://stackoverflow.com/a/9019461/185171

2: https://en.wikipedia.org/wiki/CIELAB_color_space#CIELAB

3: https://github.com/worldveil/photomosaic/blob/bb720efda11383...

In addition, in Tim's post on the python mailing list he says he developed the algorithm after reading through a number of papers on sorting[1]. So while the algorithm was developed with the goal of being practical, plenty of academic research was used to develop it.

[1]: https://mail.python.org/pipermail/python-dev/2002-July/02683...

Worth noting that ucontext is quite slow (at least on linux): https://www.boost.org/doc/libs/1_67_0/libs/context/doc/html/...

C library public headers are commonly two levels deep to so that projects using them can add the "include" directory to their header search path and in their code have #include "<libname>/header.h". It helps avoid filename clashes.

Your rotation widget is nonstandard compared to most other 3D editors. See for example Blender[1] or Maya's[2] widgets (which are both basically the same).

It's also kind of confusing as it seems like pulling one axis' rotation manipulator actually rotates about a different axis. So pulling on X rotates about Z, Y about X, and Z about Y.

[1] https://wiki.blender.org/index.php/User:Simtron/Doc:2.6/Manu... [2] https://knowledge.autodesk.com/support/maya/learn-explore/ca...

I would try unrolling 2-4 iterations of the loop. Multiple sequential loads isn't much slower than a single load, so batching your loads and stores together will let you do more arithmetic operations for each time you hit memory.

Depending on the latitude sunset might be when people just start to get off work, or it might be well after dinner. Times have more semantic meaning than just the position of the sun.

Timezones are useful because they let us answer the question "what point in the day is it at <location>." People's schedules are more or less aligned around local sunrise and sunset so knowing what time of day it is at a location important when communicating with people in that location. See https://qntm.org/abolish

China already tried getting rid of time zones, and even though it's only 5 time zones wide it caused enough issues that far western provinces use their own unofficial local time: http://www.theatlantic.com/china/archive/2013/11/china-only-...

I'm not a fan of 0 (zero) having rounded sides and O having straight sides, I think it should be the other way around.

In LAPACK you would use dgetrs to solve for x directly with the LU factorization.

Also for the 3d every direction needs to be visible by at least 2 cameras.

I don't think they're textured cubes, each larger cube is made up of a grid of individual voxels, each with a single color.

thewirecutter.com also has ads

Despite the fact that signed overflow/underflow is undefined behavior I'm pretty sure that many more bugs have resulted from unsigned underflow than signed overflow or underflow. When working with integers you're usually working with numbers relatively close to zero so it's very easy to unintentionally cross that 0 barrier. With signed integers it is much more difficult to reach those overflow/underflow limits.

How is it more difficult to control what people say on USENET than on the internet at large?

I think it's more likely because Unity has been targeting the smaller dev market since it began, while Unreal only seriously started targeting that market earlier this year with its free version.

Since you're using the auto-tuned index for FLANN it's going to build a large number of indexes to try and find the best performing one, which can take a long time. Properly tuned I would expect FLANN to be much faster (though that's expected since it doesn't do exact searches). Also benchmarking the index build time together with query times doesn't make sense to me since depending on the application you may not care how long it takes to build the index.

I would imagine the crunch is significantly lower on a game engine as opposed to a game where you are usually contractually obligated with your publisher to release a final product on a specific date.

I read the entire 5 pages and I have absolutely no idea what this company is selling

I think the simplest way to do stuff like this is to build up an iterated copy of the previous state and once done replace the old state with the new one. Iterating isn't modifying the current state, it's generating the next state from the current one.

An optimizing compiler may not be better than me at optimizing hot code paths, but my time is a very limited resource. The compiled version may only be 75% as fast as my hand optimized version, but writing that hand optimized version will likely take several times longer. Sometimes it is worth spending the extra time for that performance, but usually it is not.

Modern sports games usually need the same stats from a simulated game as they would get from a played game so simulating is often just running a stripped down version of a full game.

It's not a 1 in a million chance for each record, it's a one in a million chance of their being a collision in the entire set of 100 million records. So for every million sets of 100 million records you generate you'll get on average 1 set with a collision.