I analyzed 376 image-required middle-school math problems with multimodal LLMs.
Most people assume “no image = impossible.”
That’s not what I found.
The surprising pattern:
• Some problems were still solved correctly without the image
• Some were only solved by luck
• The hardest failures came from counting, scale, and spatial interpretation
I’m sharing the most interesting examples and mistakes next.
Follow for real data on where AI actually fails at visual math. 🧮
Behold: the ultimate farmyard chimera! 🐷📷Half bacon, half nuggets. When pigs fly... and cluck.
What should we name this glorious beast? #AIArt#HybridAnimals#WhenPigsFly
Back from the void—now rocking X Premium Plus! Grok 3’s razor-sharp AI hooked me in, and I’m here for ad-free scrolling + maxed-out replies. Dropping hotter takes soon. Who’s ready to stir things up? #XPremiumPlus
Thrilled to join X Premium Plus! 🚀 Ad-free, boosted replies, and long-form Articles are great, but Grok 3’s cutting-edge AI was the real push for me to subscribe. Ready to spark some bold convos—let’s light up X together! #XPremiumPlus
Exploring intersections of #LearningScience, AI, and online education at @WPI. Working on making math more accessible with adaptive tutors and LLM-driven alt-text for students with vision impairments. Bridging tech and education for inclusive learning! #PhDLife#EdTech#AIForGood