Sterling Crispin ๐๏ธ
@sterlingcrispin
Artist + Software Developer / Applied AI / Agent Systems / Autonomous Trading / Married to @Helen_Crispin_ / Previously AR-VR and Neurotech
๊ฐ์
August 2009
6.1K ํ๋ก์ ์ค 46.3K ํฌ
This really, really does not bode well for the wetlab idea.
Seems like thereโs a huge disconnect between these models being aligned in text VS aligned in their multimodal reasoning
GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%.
๋ ๋ณด๊ธฐ
