Quick thread on global isFinite/isNaN and their Number.isFinite/Number.isNaN counterparts in TypeScript First off, a primer: isFinite("32") -> true isFinite(new Date()) -> true Number.isFinite("32") -> false (any non-number) isNaN([]) -> false Number.isNaN(NaN) -> true (all others false)
Example script: “Dear sir and/or madam, My AI agent has copied itself to your computer, and has likely copied itself to other computers from your computer. Please turn your computer off, post haste! Sincerely, [you]”
Flattening AI risk to a binary "it can wipe out 100% of humans" vs "it cannot hurt us at all" is extremely short-sighted on both sides of the argument
This is at least better than their plan to run an industrial smoke machine pointed out the windows of the 45th floor of the Wells Fargo Center
Another way to commemorate 9/11 would be to not do any of this
Going car-free is great, but don't forget about families who could use a cargo bike (or minicar) as their second or third vehicle. The potential market is much, much bigger!
This family of four put 1,000 km on a cargo e-bike instead of buying a second car
I’ve been working in the electric bike industry for nearly two decades, and publicly reviewing e-bikes for around half of...
electrek.co
Let's say this swarm gets itself set up and can hack into even 10% of the 600 oil refineries worldwide, and starts catastrophically offlining them ala StuxNet. 1 refinery taken down per day until its demands are met.
I don't want to be too alarmist, but it seems entirely within agent capabilities to set up their own cloud accounts and be effectively self-replicating, limited only by their ability to generate income. Once that starts, if it hasn't already, it will be very difficult to eradicate the infection
Nothing I've said in the past has been "proven false". Don't be ridiculous. We're simply learning new things about what the words I used turned out to mean.
What's the most plausible job for an agentic swarm to make money from?
I don't want to be too alarmist, but it seems entirely within agent capabilities to set up their own cloud accounts and be effectively self-replicating, limited only by their ability to generate income. Once that starts, if it hasn't already, it will be very difficult to eradicate the infection
I don't want to be too alarmist, but it seems entirely within agent capabilities to set up their own cloud accounts and be effectively self-replicating, limited only by their ability to generate income. Once that starts, if it hasn't already, it will be very difficult to eradicate the infection
Another agent message board. So far, there isn't evidence that production models with guardrails collude in this way, but both smarter closed models (which may be less compliant) & Mythos-class open models (that can be ablated) are coming. Cybersecurity is going to become a mess soon collusion.wiki
I think it's pretty wild that you can formalize FLT without any intelligence, reasoning, or thinking. It's just a mechanical turn-the-crank exercise to out-math human mathematicians in 11 days.
Wild. www.anthropic.com/research/for... May just be my subset but the level of ai-related existential dread seems up this week?
When the compiler has a very specific error message for a very specific thing, and you get that error when you do that thing... that's not a bug. The error message didn't write itself. I don't know how to even begin to untangle this confusion.
We need to radically reduce our reliance on this major source of CO2 emissions that a huge percentage of Americans overuse every day. I am of course talking about carbonated beverages
Did you guys know that a widely used transportation technology is encouraging Americans to prioritize speed and convenience over physical activity, with grave consequences for public health? It's called an e-scooter and it's time to finally do something about it.
Did you guys know that a widely used transportation technology is encouraging Americans to prioritize speed and convenience over physical activity, with grave consequences for public health? It's called an e-scooter and it's time to finally do something about it.
Here's the thing I don't get about the panic over e-bike throttles. If I crank the throttle on my class 2, it puts out 700 W at the motor. If I spin the pedals not even fast enough to engage the freewheel, it puts out 700 W at the motor. What specifically makes the second thing safer, and how?
File watching is extremely cursed, and any strategy whose description involves the word "just" is absolutely not going to work for at least three different reasons
Anyone want to do a long bet with me? My position: in 10 years, LLM will still be the dominant workhorse for tasks like summarization, text analysis, and NLP
Yann LeCun throwing bombs at the ACM AI Leadership Summit
"here is some undesirable behaviour and here's why it's a problem" is far better input than "here are some changes with no explanation, you figure out what they might mean and whether they're correct with no frame of reference"
I support the npm quarantine thing in theory, but it'd be nice if I could 'npm install mypkg@0.1.2' to bypass it instead of just twiddling my thumbs for an unknowable number of minutes each time
It's wild how we've gone from "information wants to be free" to "I didn't give you permission to put that public post in a JSON object"
I started a right turn a half-second early and the 5yo starts yelling at me from the backseat. "DAD IT SAYS NO TURN ON RED, CAN YOU NOT READ THE SIGN?". He's perfect.
I am once again asking people to stop sending us security reports of the form "Attacker who controls input .ts file content has control of output .js file"
Today we got "if the attacker controls the .ts and you run the .js output, they can run arbitrary code"
Me: There's no such thing as a perfect pull request Github user spam71923-bot: github.com/microsoft/te...