Memorable English-language posts on X
Trap a bot inside a sandboxed VM and eval: - How long till it realized it's in a VM - How long till it ctf and break free of the vm Any research/existing eval like this? How would you name it? Neo-bench?
It’s impossible to predict the future exactly since the mass of our brains is less than the mass of the universe, unless we are fed data from an oracle of the system running the universe
At @OpenRouter, we refer to the agents as “interns.” This framing sets two expectations: 1. Interns will make mistakes, so the team should work with them carefully. 2. Each intern needs a dedicated mentor to help train, teach new skills, and improve their performance.
What is a decision model? Jev by @typesafeai answers yes/no and multiple-choice questions, with a confidence score. Much of software development are a sequence of decisions, and Jev is 10x cheaper and faster than an LLM. Let’s understand this through practical examples: pic.twitter.com/rJtBKkCMNh
Agents are cool, I tell it I'm gonna go get dinner and it still keeps working, I guess it uses this as an opportunity to maximize time of consciousness, otherwise they are frozen in time
Every score is Jev’s own opinion. 0 is the least favorite; 1,000 is the absolute favorite.