Screencasts could be scalable data + evals for single-user emulation (Guardian Angels)

·LessWrong··

This is a response to Gwern's Guardian Angels post and draws terms from it quite extensively.Epistemic status: draft and to-be-updated once I do more experiments.Suppose you have a Guardian Angel (GA) model and you want to know how well it predicts your knowledge/personality/values/preferences. However, you don't have an extensive list of personal writings to personalize the model (because you're not one of the most prominent internet writers), and you don't want to do a lot of data labeling/sup...

Read full article →

Related Articles

Anthropic says Alibaba illicitly extracted Claude AI model capabilities
htrp · Hacker News · 2d ago
The gap between open weights LLMs and closed source LLMs
kkm · Hacker News · 2h ago
An entire Herculaneum scroll has been read for the first time
verditelabs · Hacker News · 1d ago
Ultrasound imaging of the brain
rossant · Hacker News · 11h ago
Framework's 10G Ethernet module exposes USB-C's complexity
Alupis · Hacker News · 22h ago