原帖内容

I tested Glimpse by @ElorianAI through early access by uploading a mystery painting and asking one simple question: How many horses do you see in this painting? Glimpse broke the image into four sections, zoomed in to examine the details, and analyzed the visual clues before reaching its final answer: 5 horses. Then I tried the same image on other models, including GPT-6 Sol and Gemini Flash. While they provided direct answers, I found Glimpse’s approach more impressive because it went deeper into the visual analysis instead of simply returning a result. What makes this launch even more interesting is the team behind it. Elorian’s founders helped build Google’s Gemini and Apple’s MM-1. Now, they’re tackling a harder challenge: teaching AI to reason about what it sees. That’s what I wanted to explore firsthand. It’s not just about getting the right answer; it’s about examining the evidence, breaking down the problem, and understanding how AI reaches its conclusion. Could visual thinking be a major step toward AI that understands images and videos beyond surface-level recognition? Curious to see where @AndrewDai and the Elorian team take this.