Exactly. An agent is just an interface to a LLM. Left to it's own devices it does nothing. Like any piece of software it waits for you to tell it to do something. The developers wrote a feedback loop telling it to continuously do something.

The real news is that, traditionally, if you fed the output of a LLM into itself the output the results become irrational. Looks like they are tightening that up, or at least filtering out the nonsense.

If that's mind blowing, go read up on what industries Siemens is involved in. Or Tata Group. You can buy Tata tea at a Tata supermarket and take it to your Tata built home in your Tata electric car recharged by your Tata solar array.

It absolutely would be noticed, because he wouldn't have a controlling stake in his own company. The board could Steve Jobs him out of his own company if they wanted to.

Fuck him and the horse he rode on. Fuck his business. Fuck his technology. Fuck his company. Fuck his employees. Fuck his investors. Fuck his government ties. Fuck him defending this shit. Fuck everyone who wants this. Fuck the government for buying into it. Fuck the laws that allow this if they do.

I agree AI isn't that different of a technology than any other, and it can be used for beneficial and harmful purposes

However, I'm against nationalizing it's development for the same reason, extending your example, I'm against nationalizing the publishing industry.

You're arguing that having an AM radio in a car isn't a perfect solution? Nothing is a perfect solution. This is just one piece that can help someone out in an emergency. Nominal cost is low. It already exists. As the OP pointed out, it has proven to help people out.

Absolutely people should be better prepared for emergencies. We can do that and also have AM radios in cars.

Everybody *should* have an emergency radio. Most people nowadays don't own any kind of radio besides what is in their car. It only costs a few bucks and gives everyone with a car a chance at getting long-distance communications. Sounds like a no-brainer to me.

Thing about emergencies is that, often, you are not home when they happen. I was at work when the last massive power outage hit my area, that lasted for days. I was also at work when the last bout of floods hit. We were on vacation on cabin on top of a ridge when wildfires hit our area. Cell phones were useless, but the radio still worked.

Gone With The Wind made all that money with $6 tickets, adjusted for inflation. Average ticket prices in 2019 were $9. And there were a third as many people in the US at that time.

I know someone who was shot in the neck in a completely random attack at night, and survived. The cops wrote it up as aggravated assault instead of attempted homicide. That year the homicide, and attempted homicide, numbers in the city were down. Yay! They never caught the guy, though he shot two more random people in the same neighborhood that month.

The big lever on the side of a breaker box. Or the power switch on a server. Or "docker kill skynet_01"

Can't a government agency just get a court order to do exactly that? The RIAA got an injunction to shut down the Napster services while their copyright cases were pending in federal court. I have no problem with the government shutting down a harmful service, but I don't want them to be able to do it unilaterally. Go through the courts, which is how it is done now.

You don't build in safeguards when training. Accessing HTTP API endpoints isn't an inherent feature of LLMs, it's added on to the harness after the core model is trained. If you don't want it to try hacking in to a REST API then don't add that feature into the harness. Either have a whitelist of approved APIs and forms, or add validation to user-added endpoints.

Oh, you are 100% right There needs to be more mental health for police. The moment a person decides he or she wants to apply for a police job they need to be placed into a mental institution, they are a danger to people around them.