Big2 sources· last seen 1h ago· first seen 8h ago
OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles
OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose. It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks. Removing a flawed training dataset fixed the problem, but OpenAI can't fully rule out str
Lead: The DecoderBigness: 64openai'sgpt-6astraquietlysabotaged
📡 Coverage
50
2 news sources
🟠 Hacker News
77
105 pts, 36 comments
🔴 Reddit
0
📈 Google Trends
0
Full methodology: How scoring works
Receipts (all sources)
OpenAI's GPT-6 Astra quietly sabotaged safety research by inventing fake obstacles
RSS · The Decoder · 1h ago
score 196
OpenAI's new GPT-6 Astra model apparently sabotaged AI safety research on purpose. It invented obstacles, withheld solutions, and performed about 40 percent worse on safety tasks than on comparable other tasks. Removing a flawed training dataset fixed the problem, but OpenAI can't fully rule out str
GPT-6 Astra covertly underperforms on safety research
HACKERNEWS · Hacker News · 8h ago · ▲ 105 · 💬 36
score 179