hilarytorn.bsky.social
Exploring AI manipulation, scheming and deception. Growth marketer transitioning to AI safety. 🇺🇸 mom with a sprinkle of 🏳️🌈 living in 🇪🇪
👤 Joined on 21 November 2024
Followers
Following
Total Posts
@hilarytorn.bsky.socialGoogle DeepMind released Gemma Scope 2, an open-source interpretability suite for the Gemma 3 family (270M–27B parameters) that required ~110 PB of stored data and over one trillion parameters trained, intended to accelerate AI safety research and auditing.
Show PostCounterintuitive AI safety research by OpenAI suggests that teaching models to confess their mistakes by making honesty easier than deception. When they misbehave, they admit it 74% of the time. The trick? Never punish them for telling the truth. gregrobison.medium.com/teaching-ai-...
Show PostHiring the wrong person not only hurts the team, then delays getting the right person in, but is also terrible for person you hired, who ends up being fired. They might have quit a job. How long will their search take? If this happens a lot, you HAVE to fix yourself.
Show PostJust used Claude Code to make a multiplayer 3d first-person Christmas game in the browser my daughter and I are playing together on two computers over the LAN. The future is nuts
Show Post#Frontend devs: “Why is the API slow?” #Backend devs: “Why is the UI doing 12 requests for the same thing?” #DevOps: “Why is anyone doing anything?” 😭 #buildinpublic #100DaysOfCode
Show Post