BREAKING
Ex-DeepMind Researcher Launches Elorian
Frontier models reason like preschoolers
What breaks visual reasoning
1
Draw pelican SVG
↓
2
Count fingers
↓
3
Maze navigation
↓
4
Count 10+ objects
0
M
seed raised
0
M
valuation
Vision-native, not text-first
Existing models
text-first
●
Convert images to text
●
Weak spatial reasoning
Elorian
vision-native
●
Reason in visual domain
●
Physical-world tasks
No model released yet
AI NEWS BLITZ
A former Google DeepMind researcher just launched a new AI lab focused on visual reasoning.