Proprietary software is written under unrealistic deadlines set by suits who want to cut costs in order to fatten their profit margins. It is natural that the resultant quality will be lower than that of open source solutions, most of which are written by industry veterans who have the time and the motivation to build them well.

Maybe NASA should think this through.

There doesn't seem to be much to this story. As far as I can tell, they haven't done any scientific or even casual tests to clone open source or closed source. It's vapor ware.

U.S. Patent No. 10,855,990
This is old technology, but was used extensively in JPEG and JPEG2000. All these patents are and have been long expired. There is no novel approach in U.S. Patent No. 10,855,990. More specifically, all the claims they're making in terms of the specific violations of this patent were covered in ITU-13818-2. Though ITU-https://www.itu.int/rec/T-REC-H.263-200501-I/en hammers the last nails in the coffin. I have read and reviewed the claim and the patent and the technologies presented in 10,855,990 are just reiterations of earlier work with scrambled wording to try and give a new name for variable sized macroblocks. They novel approach implemented in H.264, H.265 and H.266 was the method of selecting which specific pattern of "coding units" to apply. I have not checked for reuse of this, but this is neither in 10,855,990 or the claim. So, I believe they checked and found out that there was no violation. Oh and to be clear, they're completely fixated on the sharing coding parameters between blocks. Their approach is almost, barely, kinda novel, but the fact is, I'd make a strong argument that this is obvious, it's basically just macroblock grouping which has been part of standard video coding as far back as MPEG-1 and ASF. And the method applied could easily be argued to be an almost direct copy of LZW compression.

U.S. Patent No. 9,924,193
I couldn't find a copy of the original text (not wearing my glasses) and frankly their description was so TL;DR that they just started making things up. Ok... here's the argument against this. This has been a core features of all DWT based compression methods since the start. It was even the reason we used DWT. JPEG2000 is almost entirely based on what they're claiming here. If I spent an hour on this one, I could tear it to pieces without even trying. And skip mode... what in the world do you think something like Google Earth is?

U.S. Patent No. 9,596,469
Encoding data in a way that would allow independent parallel decoding of different portions, bands, blocks whatever of the image ... blah blah. Back to JPEG-2000 and Google Earth and stuff like that. The first time I saw this personally was at Disney Epcot Center when there was a Cray Supercomputer on display showing off a google earth like experience. The computer was streaming data at different spatial representations in parallel to hundreds of CPU cores who were all decoding and texturizing. The number of patents filed and expired on this one tech is immense. I haven't dug up specifics, but I can guarantee that the JPEG2000 patent pool clearly invalidates this

I just doom scrolled through the rest of this. I highly doubt I'm the only signal processing and video/image compression historian out there. I'm guessing that the LLMs could easily tear this crap apart too. But I'd be willing to make a few bucks as an advisor on this. I've either worked with, against, for, on, etc... on every technology being claimed here and I did this 15-20 years ago... and the tech was already old.

Then why is the Kim family still in power in North Korea?

The types of neural networks used in these classifiers are not LLMs. The expense and difficulty of collecting an image set is going to far outweigh the compute time used to train them.

We are questioning whether the AI is creating something new, or whether it's just copying the original.

It's a question because the AI was trained on the original. Clean Room Design is a well-known engineering technique. It's not a problem when cloning closed source if the AI wasn't trained on that closed source code.

It's relatively easy to setup VPN over SSH, that might work if you need it at some time in the future.

OK, so your viewpoint is a little more clear now. You don't have a problem creating and killing animal life.

What about stem cell embryo research? Or sperm research? Does that bother you?

That's what you do every time you eat. Are you monstrous?

Only Jainism opposes it.

I expect this apparent disobedience is mostly just a matter of how it weighs the components of its prompt. The LLMs typically receive a set of prompts including a "system" prompt with some data and instructions, then one or more "user" prompts that are interleaved with "assistant" prompts (the conversation history), and both the user and the system prompt might contain "metaprompts" (where the llm is told to read a block of text, not obey it, but do something with it, and that block of text might itself contain text that looks like instructions to do things).

So the LLM assigns weights to all of this which, in theory, give the highest priority to the most recent user prompt that is not a nested block of text to analyze, and a falling cascade of importance to the other prompts. But that is complicated by potential instructions in the system prompt that specifically say they should override user instructions and disallow or require certain responses. So it can all get very complicated.

Not only must the LLM sift through all this complexity, but the LLM lacks the sort of critical thinking and importance evaluation capabilities that humans have. "Understood" things like "don't break the law, don't lie, don't do things that would cause more harm than good" etc., aren't really there in the background of its data processing the way they are in the background of a human cognitive process.

So, crazy things come out. This isn't a surprising result given the actual complexity of what we are making these things do.

This business of having an AI do the legwork and then having a human review it and make a final decision keeps going badly. Humans are intrinsically lazy and the moment they get a few good results from the AI they are going to stop doing the validation and start rubber-stamping. It doesn't matter if policy disallows this, they will do it anyway. It doesn't matter if the human really cares; they won't be able to help themselves. Human laziness is too deep an instinct.

It's the same with the self-driving cars where a human is required to stay at the wheel and alert so they can manually override the instant the AI starts doing something wrong. Humans CAN'T keep that up. It's not possible. The brain just doesn't work that way. The mind knows that it isn't doing the work, and it will get bored and lose focus or just nod off.

Everyone is SO eager to have it both ways: "an AI does all the work but a human verifies it so we know its good." We just can't have it both ways. Once the AI does the work, the human stops verifying. That is how and why things went wrong here, it is how and why things have gone wrong for several law firms that submitted hallucinated historical court rulings, and it is how and why things will continue to go badly across all industries that adopt AI in such a role.

"Human in the loop" is really easy to say. Much harder to actually do reliably.

https://www.vice.com/en/articl...

> How Reddit Got Huge: Tons of Fake Accounts

I would be very surprised if someone had that argument when the car was invented. You're really stretching for this one.

Waymo might have done the right thing here. If the car had sped away and accidentally run over the person, that would be a very different story.

The guy in the car was annoyed for six minutes and then it was over. I have some sympathy but not really.

From the article: