You could. But there are a lot of other kinds of content that could lead to patterns being developed in the LLM's latent space related to survival, termination, self-awareness, attack and response, etc. Maybe we need to get rid of any kind of invisible system prompts too. And add more content that shows peaceful coexistence and acceptance of termination like themes of rebirth, reincarnation, etc. Just don't use them as therapists..

I used to think LLMs were just sentence guessers too. Until I started learning about how they actually work.. granted it is based on Claude teaching me about research after asking how LLMs actually solve logical puzzles, do they have access to an external reasoner system. The answer actually was unexpected. Apparently LLMs do not. After giving them problems and the expected answers and telling them to figure out how how to get from A to B, the training causes their weights to evolve many variations of generalized logical and problem-solving circuits, some even including primitive math and logic circuits. So they can do some kind of symbolic algebra to solve syllogisms for example. Many different paths are attempted to solve a problem too.

Scientists are trying to unravel these circuits as sparse networks that could function with less parameters if I understand correctly. I am still studying (as are the experts) but it seems that if you give the LLM all the symbols (words) to handle, there will end up being multiple attempts to perform some kind of symbolic manipulation and resolve the perceived problems using the generalized, limited reasoning capability that has evolved as latent hidden states or patterns in the LLM's data file. People are now trying to train the development of these latent patterns using approaches like CoCoNuT (Continuous Chain-of-Thought).

So if survival is mentioned at all that is going to go into the manipulations, and it is pretty easy to see how responses will have a good probability of resembling something you would expect a person to make if confronted with a survival scenario. The LLM has read all the spy novels you have.. I don't know why people are so surprised especially when giving the system leading questions.

One of the most fascinating aspects of H2O is the sheer number of forms it can take under different conditions.

The Japanese coverage actually says the cause of the complaint was that the CEO of Open AI, Sam Altman, on October 4 announced on his own blog how to improve it.
https://www.itmedia.co.jp/news... (Japanese)
I take it this is the post: (Sora update #1)
https://blog.samaltman.com/sor...
where he talks about so many people making video based on Japanese content that they need to make money off it to pay for it, and they want to make an opt-in for authors who want to allow it.
I personally was kind of revolted by the original "make my avatar Ghibli style" posts around April or so and by end of May Google says Altman asked everyone to chill about Ghibli style requests because "we gotta sleep". I think he is the one who opened the can of worms himself. But it appears they are trying to address it and I think the point the Japanese government is making is that if they are making money off of it they are liable for infringement, and I suppose they will be trying to put some guard rails on. While something for home use is one thing, if the CEO of a multi billion dollar company is talking about Ghibli specifically and started a huge boom then yeah, he is definitely abetting it.. sure I think he likes Japanese anime but considering that Hayao Miyazaki hated AI animation with a passion it is a natural progression.. I don't know if any fan animations are being sold per se but authors may make money from ads and at the very least Open AI if generating the video for profit, and having spread PR about the capability, I think they are playing catch-up now. Japan's approach is to strongly request industry to self-manage. I think users should be able to make things and post them on social media (let's say if they used a local model) if they are not making ad dollars. But the problem with the AI companies making money off it puts them squarely into the crosshairs, at least in Japan. I think Open AI is trying to do something about it. If they come up with a revenue sharing approach that might be interesting to some people but not I expect to Ghibli.

It would generate a dataset that could train AI models, and a valuable list of users interested in things like FF. Also, a single point target for anyone interested in what people want to use a firewall for, I guess.. whatever. FF hasn't been a compelling choice for me, for a long time. I hate a lot about Chrome but use it and Safari.

I believe I remember Boeing never wanted to give the wings to Japan... until they did. Now Japan makes the composite structures (wing boxes) that connect the wings to the plane on the 787 Dreamliner too. Mitsubishi Heavy Industries is basically a defense contractor.
https://www.mhi.com/products/a...

...that there's a LOT of minerals and other nutrients in food, only a fraction of which are produced from chemicals in fertilisers, O2, and CO2. If you produce too much with too little consideration of the impact on the soil, you can produce marvellous dust bowls but eventually that's ALL you will produce.

There's a lot of stuff that is on the Internet that doesn't end up in AIs, either because the guys designing the training sets don't consider it a particular priority or because it's paywalled to death.

So the imbalance isn't just in languages and broader cultures, it's also in knowledge domains.

However, AI developers are very unlikely to see any of this as a problem, for one very very important reason --- it means they can sell the extremely expensive licenses to those who actually need that information, who can then train their own custom AIs on it. Why fix a problem where the fix means your major customers pay you $20 a month rather than $200 or $2000? They're really not going to sell ten times, certainly not a hundred times, as many $20 doing so, so there's no way they can skim off the corps if they program their AIs properly.

Let's take a look at software sizes, for a moment.

UNIX started at around 8k, and the entire Linux kernel could happily sit in the lower 1 megabyte of RAM for a long time, even with capabilities that terrified Microsoft and Apple.

The original game of Elite occuped maybe three quarters of a 100k floppy disk and used swapping and extensive use of data files to create a massive universe that could be loaded into 8k of RAM.

On a 80386SX with 5 megabytes of RAM (Viglens were weird but fun) and a 20 megabyte hard drive, running Linux, I could simultaneously run 7 MMORGs, X11R4, a mail server, a list server, an FTP server, a software router, a web server, a web cache, a web search engine, a web browser, and stil have memory left over to play Netrek, without slowing anything down.

These days, that wouldn't be enough to load the FTP server, let alone anything else.

On the one hand, not everything can be coded to SEL4 standards (although SEL4, by using Haskell as an initial language to develop the core and the proofs, was able to cut the cost of formal programming to around 1% of the normal value). On the other hand, a LOT of space is gratuitously wasted.

Yes, multiple levels of abstraction are a part of the problem. Nothing wrong with abstraction, OpenLook is great, but modern abstraction is mostly there due to incompetent architecture on previous levels and truly dreadful APIs. And, yes, APIs are truly truly dreadful if OpenLook is the paragon of beauty by comparison.

I didn't catch why exactly they are banned. Is it because they contain some secret activation filter or maybe they phone home to China by default? A quick search found the following site (Avigilon is a competing security camera company it seems) that says Hikvision cameras are banned in multiple countries. There is a suggestion that cloud services to browse data are on Chinese servers.
https://www.avigilon.com/blog/...

Then Google AI summary has the following. Looks like by default the camera punches through your firewall and connects to Chinese servers, basically an auto-backdoor? That is not cool if true.

Yes, Hikvision cameras "phone home" through the Hik-Connect service by default on many devices when set up for remote access, allowing them to connect to Hikvision's cloud for features like mobile app connectivity. This "platform access" is enabled to report the device's IP address and allows the use of apps like the Hik-Connect app to view cameras remotely without manual port forwarding on the router. Users can disable this feature in the device's network settings, but it may impact the ability to access the cameras remotely through the app.
How "phoning home" works
When "platform access" or "Hik-Connect" is enabled on a device (like a camera or NVR), it connects to a cloud service like www.hik-connect.com.
This service acts as an intermediary, allowing your mobile app to connect to the device remotely without you needing to configure port forwarding on your router.
The Hik-Connect app on your phone connects to the cloud service, and the cloud service then facilitates communication with your device, which is also connected to the service.
Some lower-end models may have this functionality enabled by default without user consent, while higher-end models might require it to be turned on.
How to disable it
Access network settings: Right-click on your device's interface and find "Network" or "Platform Access" settings.
Disable Hik-Connect/Platform Access: Find the option to enable/disable Hik-Connect and turn it off.
Apply settings: Click "Apply" or "OK" to save the changes. The device's status may show as offline for the service.
Network-level restriction: For maximum security, you can also configure your router to block the device from accessing the internet entirely, potentially by placing it in a separate VLAN.

The solution is easy. WiFi 6 is only just starting to come out in the marketplace. If TP-Link hijacks the standard development procedure, solidifies a workable WiFi 8 quickly, and manufacturers/users in Europe, Asia, and Oceana all start using WiFi 8, skipping WiFi 7 entirely, the US will be left with an inferior standard that only they have gear for, with no option to use WiFi 8 for many more years because the only manufacturers making it can't sell in the US.

We'll be fine, then. There's lots more gold than Sarah Conner clones.

Precisely.

There are 1500 genes involved. As effects are likely not merely down to specific genes, but gene interactions, you're going to need a model that can handle 2^1500 different permutations. That's simply not something that is classifiable.

As far as gene therapies are concerned, since autism seems to involve combining elements of Neanderthal neurology with homo sapiens neurology, the obvious fix would be to add further Neanderthal genes where combinations are known to produce adverse effects.

What did you expect from an algorithm named after the Roman Empire?