Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

#ai #deeplearning #computerscience #machinelearning

Think you beat question-answering? Try this tougher AI challenge

Think you can outsmart a machine at simple school science? A new public test throws a real head-scratcher at researchers.
It collects real, human-written science questions kids see in class, and then picks the ones that ordinary search or word-matching systems kept getting wrong.
That makes a hard challenge — one that needs deeper reasoning, not just finding words.
Teams tried strong models from other exams, but most methods still did about the same as a random guess.
The project also shares a big background set — about 14M sentences of science text — so builders can try new ideas.
This is about plain, grade-school science turned into a test of real thinking.
If you build models, this is a clear invite: bring your best, see if your model can do better.
Try it, learn from the failures, and maybe teach machines to really understand the why, not just the what.
Good luck — its harder than it looks, but exciting.