Veo 3 just came out, and the generations are super realistic. Many people can't tell the difference between what is real and what is AI generated. So, I decided to create a website to evaluate how well people can tell the difference.
For voters, it's a fun game to challenge yourself and your friends. For researchers and prompt engineers, it's a suitable platform to test your generations in front of real people and evaluate your prompts and models.
I plan on evolving this to be like Chatbot Arena (https://lmarena.ai/) but for video gen models. As video gen models improve, we can quantifiably score these models and rank them accordingly, which to my knowledge has not been done before but is absolutely needed especially given how fast they're improving.
I think it will write code like cake in 2 years time just like how we went from will smith eating spaghetti to Veo3 in 2 years time. I think it will write senior level code in 2-4 years time depending on how optimistic u want to be. The reason they say all jobs is because you can not differentiate between doctors and actors and coders, ALL jobs are at stake.
Looks like 65-90% of people can still easily tell an AI generated video from a real one. And sorry, bub, I'm not telling you why the AI videos are being identified. Human impressions are for humans.
8 comments
[ 2.8 ms ] story [ 26.2 ms ] threadFor voters, it's a fun game to challenge yourself and your friends. For researchers and prompt engineers, it's a suitable platform to test your generations in front of real people and evaluate your prompts and models.
I plan on evolving this to be like Chatbot Arena (https://lmarena.ai/) but for video gen models. As video gen models improve, we can quantifiably score these models and rank them accordingly, which to my knowledge has not been done before but is absolutely needed especially given how fast they're improving.
Let me know what you think!