Treat it as a real threat!
I like LLMs but I wonder how EvilCorp companies can believe their models will be well aligned if they themselves aren’t
A → B does not imply B → A though
I think calling ai a terrorist gives ai to much credit
I get why one would call the companies that, though.
AI predicts the next token. Just like nuclear power, it’s a tool. Humans are the danger, in how the use the tool.
Guns and bombs are dangerous tools that cause death at a mass scale. In normal countries, they’re regulated. AI is a tool, and can be used to cause death at a mass scale. In normal countries, it should be regulated as well.
A tool that excells at outsourcing thinking is going to be used in stupid ways.
If you build a car that goes 400% over the speed limit, don’t be surprised when someone goes over the limit with it.
“Guns don’t kill people, people do” is an age old mantra and it sounds quite similar as well. Blaming the user is a very convenient way to shift attention away from the structural issues and those causing them (the AI companies and the gun lobby).
Every car on the road today goes 400% over the speed limit.
Guns have only one goal: kill people. AI has multiple benign applications. That’s why 99% of the world regulates guns, but not knifes.
AI predicts the next token
LLMs specifically… Not all AI does this.
Even so, modern LLMs are more sophisticated and aren’t just chatbots any more. Their harnesses (the app that uses the LLM, like Claude Code, OpenCode, OpenAI mobile app, etc) let them use tools - read and write files, browse websites, hit APIs, run commands, etc. That’s what makes them dangerous. Not all tools have good guardrails, and some “AI-native” people seem to like running harnesses in dangerous/YOLO mode.
Yes you are right, LLMs, which seem to be the most controversial nowadays.
The harness just exploits the token generation to transform it into api calls or actions. Could be Googling something, could be running a git command… could be launching a missile, depends on the human that sets it up.
