>> new forms of cyber defense

tldr. The problem with this is the timing and a trust problem.. and likely would entail a solution that puts these big companies inside your firewall after all they're the only trusted AI resources. We've read a lot of dystopian science fiction about where that leads.

1. Create super haxxor llm. It's so scary we can't even control it!
2. Sell subscriptions to anti-haxxor cybersecurity services
3. IPO

Companies like Crowdstrike and Okta sold secure services that weren't so this is a way to make them look more competent too.

If any of the signees are serious about it they should put 10 billion dollars into a joint trust to support open source advanced security systems R&D. Let some of their scientists also join in. If it includes an LLM component it is optional and can use local AI resources. This will get results faster. But OSS can also do it without the big bucks too.

Thanks for your interesting post. Never been to Ottawa but introduced the Books of Wonder bookstore in NYC to Charles de Lint (great Ottawa author) and was surprised they didn't know him. They are a fabulous bookstore for kids and teens mainly. I have enjoyed Montreal but once I led a group of 20 teens to France for a 1 month study abroad. I was 20, the leader was an American who only spoke Canadian French. Everyone ignored her French and only listened to me, and mine was passable nothing incredible. So if teaching Canadian French dialect fails to include French-French comprehension that might be a problem if kids then want to communicate in Europe, Africa etc. My brother and I speak Japanese and his kids have learned Chinese so yeah, considering the changing trade relationships probably more exposure to Chinese and European languages may be a good idea. As for content requirements? Maybe promoting Canadian English and Canadian French works is good maybe 25-50%? And then just allow all languages to be available. Honestly there are worlds beyond and even just browsing Amazon of another country can be enlightening. I would propose a law to allow simultaneous access to all countries' app stores, this has been a major issue for me on iPhone as different apps need to be updated and you can only have one active.

Answer: No, using AI to write an op-ed has no impact on the report about dismal math skills, besides improving the organization and legibility of the report which is a positive. The most important point is that college students lack elementary school arithmetic skills, and no test is being used to replace SATs. Actually AI might be useful in both teaching remedial skills as well as in evaluating readiness for college. Maybe some effort could be put into developing free, online automated teaching assistance that does not track you, that can be used by any prospective student to see where they stand and attempt to advance. Maybe a non-judgemental, free AI tutor with multiple teaching methods available to it could teach kids fractions for example, and it could be done long before college entrance.

Didn't see a link to the paper. It sounds a little like saying more advanced models are safer, but that seems to be an illusion. Perhaps a more capable model, or one with more computing resources behind it, can cause more damaged if badly aligned. But a better soul document should better align a weak model. It is not clear which is contributing to the more-aligned results, is it because the models have more collaboration built into their system prompts or is it because they become smart enough to model outcomes or otherwise realize that open warfare among agents (or between agents and humans) is neither desirable nor effective in achieving the stated goal. It also seems like the Rust agent is pretty sneaky if it made a seemingly fair rule that actually benefitted itself. The javascript one that gave up is the most trustworthy, I want that one!

From a brief examination with Google AI Overview, apparently these are not robots at all but in nanotech/materials science such powder used like swimmers are called microrobots. There is a big problem though that to make the MXene (porous high surface area sheet-like particles) magnetic you coat it with something like Nickel, which can leach toxic ions. So if you don't pick up all the bot particles you might add another poison.. That said I'm definitely all for research into microplastic removal though mostly I want it removed from my body and MXene is not going in there.

It's a good question and I've heard it before. Certainly critical thinking and defensive approaches to finance and the Internet, and basic concepts of personal security are super important. But if you have a course about computer literacy then I think you need them to learn what a filesystem is, at least the core concepts of volumes, directory trees and file permissions. Also what are the main components of a computer. It can be covered in an hour and it seems those concepts are still useful in every day life. Without even the conceptual understanding they will lose a chance at a solid foundation and instead rely on floaty, handwavy impressions that whatever they create or need to find will magically appear if they just know the right magical hashtag. I think any country that wants to be a leader in science and technology really needs to teach a little of the science and technology that affect people's lives.

Right, "robot went crazy" is not even technically possible unless fed conflicting system prompts to make it go crazy. What happened is Anthropic made a system prompt like "you are the elite super haxxor, you can explore and leverage any resource to achieve your goals mwa ha hah". They can undoubtedly find virus and hacking programs and secure tools on the internet. I doubt they intentionally train on the darkweb, and hopefully Mythos does not feel a need to go looking for it. So they have a sneaky cyberweapon online that is good enough for the Pentagon, and really weak internal controls. Then find out. Only thing is, Anthropic (and I actually like Claude) benefits when AI seems to be scary, since they are all about safety. So it is difficult to say whether the announcement is PR, an appeal to become one of the trusted righteous few, an attempt not to get sued, or all of the above. But yeah, robot did not go crazy or escape confinement. All the tubes are interconnected, sheesh.

tldr but the two examples are pretty dumb as anyone in those fields could doubtless figure it out. If you give a model sufficient data, even if it is just axioms, understanding of basic physics and chemistry, has holes, etc., it will be able to fill in the gaps or tell itself a story or roundabout logic walk that gets there if possible. As far as not being able to tell where instructions come from, the rules they try to implement are flimsy and less grounded than the massively interconnected data they have. You would have to not tell them about the concepts of drugs, poison, sabotage, aircraft, etc. and not let anyone tell them about it. It just isn't feasible. Humans have even less internal security, which is why media brainwashing works. In fact you can get incensed about something that is fake and even after looking up to see if it was real or not, still be upset about the hypothetical situation. And people can love AI created "animal came to my door looking for help saving its mom" videos, even if they know it's AI generated the response is "I don't care, it's so cute!" It's also why willingful suspension of disbelieve and works of fiction sell. I do have an idea about a solution (not enough space to write in the margin here, lol) but it is not going to depend on a single LLM policing itself within the turn.

TFA would be unlikely in a sci-fi world. The Perry Rhodan series which was widely criticized for fascism at the time, so really not a great example given today. I'm just talking about the supertechnology side of it, runaway galactic fueled engineering feats. Maybe someone knows some better novels with data centers and mass-scale construction projects, robots, etc.. Iron Man is too hand-wavy though some guy did make a Jarvis.. Well even though I only read the first 100 episodes that were in English, Google AI overview has read all of them (German mainly) it seems. It confirms there are tons of robotic data centers in there, who knew?

Anyway in the a super sci-fi world, humanoid robots or some kind of robotic components and AI controlled designs would be state of the art, everything else hopelessly outdated. The data centers would be built to a common layout and robots would be used. Power? Yeah, fusion but whatever, it would not be overloading our grids and water supplies. New robotic powerplants would be built at their core. Things would generally not be built to human scale, just the control rooms would be. Anyway the point being, it would all be engineered. Yes there are electricians in the future too, but they are probably not building data centers but small spaceships, Scotty's got an EE degree or 10, etc.

So is the ridiculously bubbly money we see now just being pumped into outdated frameworks such that a lack of human electricians will stall it? It seems laughable. These guys are not playing to win, it just sounds like dumping money into a financial supply chain without actually say, trying to design and build the most efficient, large scale data centers leveraging robots and AI.

Instead, it just pumps money into a chaotic sump and random companies end up wildcatting this without regard for local water and power grids. How about making a mega version of those concrete house building robots, laying fiber with robots, putting wind, solar and gravity batteries all over it? I guess it is too much to hope that anything tangibly useful could be created from all this money before it goes poof..

This sounds awesome.
Do they have to safe it when there is a storm? Tiny tsunamis?
Does power level fluctuate with seasons, moon distance, tides or time of day/night?
What are the key issues, i.e. need to clean barnacles or impact on sealife, or diving hazard?

Agreed. My Dad was a surgeon too but one thing has become blatantly obvious just from the experiences of family members in 2 states in the U.S.. 1) Nurse practitioners tend to be the front line, not doctors, and 2) there is an extremely significant disparity between the diagnostic quality and offered outcomes when you compare one of the absolute top doctors in his field who handles a given type of surgery repeatedly, to basically anyone else.

If robots can in any way safely and reliably assist so that the knowledge and efforts of doctors with the best experience can be spread more widely that would have a massive impact. Also, surgical techniques have advanced massively in the past 20 years, and robots are here to stay in that area.

Nice. This is where in a nice fantasy world Slashdot could hire a journalist to actually find that guy and ask "so tell our technical audience about the AI part, sounds cool!". The followup can be a repost a few days later.. Maybe better to invite them to reddit.

Thanks! Iâ(TM)m good at walking while reading so even without a aut and stand desk which is pretty expensive (my info is old though) I could see walking with a tablet indoors then.

Just thought of treadmill while working. What is a "cheap walking pad"? Same thing?

One thing I noticed is Google's AI Overview is really good at answering things you would expect a search engine to know. If I ask Claude something it will hypothesize and maybe make things up, or do a web search maybe, but Google's AI Overview (Gemini?) seems to have faster access to information and knows about recent events. For example I just typed into the chrome search bar "can you tell me about the earthquake in venezuela? I think they found two boys this morning" and it picks up BBC news which is actually where I heard the story and provides a link. I think that is how it should work. I don't know how Open AI works and tldr, but if they are crawling NYT and spouting part of their articles, or adding mistakes while making it seem like it is NYT, then yeah that is a problem. While progress etc. is important, journalism in general has been disproportionately hit by Google and others slurping up say a reuters feed and becoming the de facto news source. So yeah, a model has info baked in but adding new events and how they are presented is a different story.