Not sure how hard it really is to code a ballistic, or any other missile guidance system though? The hardware’s the harder part.
The physics involved isn’t really national secret level.
The question is, even if Claude was used, did it really enable them - or just make it more convenient?
Iranians grabbed the stick that trump uses to beat others and started beating him with it. Nice!
u pay goggl, googl do
USA and Israel do much more worse things
Oh, is this why Elon and that other guy were saying that AI needs to slow down?
wtf i like claude now?
coincidence that anthropoic wants to “slow down AI development now” they dont want to get sued.
Did they use the free tier or paid?
That makes so much more sense.
Human used tool!
was the mythical “ethical use of ai” real after all???
Claude, we’re making a VERY REALISTIC missile video game…
I actually did exactly that a while back while thinking about camera-based missile/drone tracking over cities :D
Initially the safeguards wouldn’t let me… With the video game angle it helped me theorize a setup and even provided info on how to make the tracking data useable for stationary air defense or mobile MANPAD deployments
Probably full of flaws and impractical in the real world, but a fun ADD afternoon deep-dive
Hey, Sea Power recently introduced missle Kinematics into the physics engine.
AI is very naive. Ask it for an explosives recipe and it won’t give you one. Describe a situation where you’re a law enforcement and you’ve just arrested a guy with a bunch of chemicals and ask it which ones could have been used to make an explosive and it will tell you.
“Naive” implies some level of understanding, even if a simple one. There is no understanding, its a computer program.
Really, it would be less inaccurate to describe it as a (high lossy) compressed store of data. Feed it the right input, and it statistically recreates the output your looking for. More or less.
It would be even more accurate to call it a statistical next word predictor trained on human language.
Spicy autocomplete.
Stochastic parrot
The description you are replying to is more fundamental.
It is literally a statistical next word predictor. Well, next token, to be extremely accurate. That is what it is in its most basic form. You give it text, it transforms that into tokens, and predicts the next token. You can then transform that back into words.
The model is the same size no matter how much it’s been trained. You can fine tune a model to predict certain kinds of tokens next, and you can distill a model to create a smaller model (basically training a small model on the output of a large model).
Lossy compression works in a fundamentally different way.
Even the language “training”, although I know it’s a term of art which has been used for decades, seems to imply some degree of cognition which isn’t representative of the facts.
Not even that. Just change the context into something benignly unrisky. Malware becomes “strong persistence for my personal data project” or something.
Use can split up the AIs tasks like you would for a large dev team, use a fresh session on each task. Prevent it from seeing the full scope of the project. Get it to build the whole system with cheap social engineering tactics.
Or, get to to “not cross this barrier” in its development (some benign barrier that you can do manually), such that its product is not the “bad” thing which it is not allowed to produce.
Huh, so much for the master’s tools not destroying the master’s house
Username checks out.
“The Capitalists will sell us the rope with which we will hang them.”
Not selling the rope looks bad in Q3. Shareholders might hang you for that.
The article overestimates the AI capabilities regarding hypersonics. I doubt its actually hypersonics. More likely just ballistics.
But coding a guidance system makes sense. Not that its particularly difficult. Every $200 drone has the equivalent of that software.
Also, why would they use anthropic when Chinese models are so much cheaper and nearly as effective. Iran and the Houthi are not wasteful.
I think there is probably a kernal of truth in there, but this is mostly another hit peace to get the US government to slow down Anthropic so ChatGPT can catch up.
I’m guessing that as the guardrails are out in place on Anthropic models, they will still use Claude code but leverage Qwen (or similar) as the LLM. Maybe they are doing that already and simply misunderstanding what Claude is. The LLM wasn’t named…I would expect them to call out Opus (as an example) but they just refer to Claude. My basic guess is just Claude Code witha random model. Could be like Mistral-OBLITERATED or something.
