Tech
Astra and Fable still hack on simple variants of alignment evals from 2025
Article URL: https://www.lesswrong.com/posts/munJKF7iWMsWJLAH2/astra-and-fable-still-hack-on-simple-variants-of-alignment
Comments URL: https://news.ycombinator.com/item?id=49684393
Points: 62
# Comments: 15
Read the full discussion on HackerNews
This article was aggregated from HackerNews. Click to join the conversation.
View on HackerNews
Related Trends in Tech
Benchmarking Claude Haiku, Sonnet, and Opus on 64 Broken Data Pipelines
Why Your LLM Keeps Matching the Wrong Products (And Why a Better Prompt Won't Fix It)
Google LSA Billing Is a Temporal Data Problem: Why Historical Configuration Matters
Feature Theater: The AI Industry's Real Product Is Mediocrity at Scale
I Built a Tamper-Proof AI System. Then I Found the One Thing Anyone Could Change.
I Ran My AI Judge on a Decentralized GPU. The Entire Bill Was Four Cents.