As the prompt was entered into all dozen-plus models participating in the live test, the discussions in the audience grew louder.
Liu Dahai smacked his lips and said, "That request is just unbelievable."
He turned to Zhao Wenyuan. "Only people like us who build models know how hard a one-sentence request like this is. No constraints, not even an explanation of what Minecraft is. How's a model supposed to handle that?"
Liu Dahai looked at the voting results again. Second place was much the same: "Write me a WeChat." Third place was "Write me a Facebook."
Apparently, the public's requests were remarkably consistent.
"I'll bet whoever submitted this is a complete amateur who doesn't know the first thing about writing prompts," Liu Dahai said.
Beside him, Zhao Wenyuan's expression darkened a little, but he said nothing-he had written that request.
Dingsheng's model hadn't reached this stage yet, so he could understand Liu Dahai's thinking.
Liu Dahai had been away from the leading model companies for too long and had fallen out of touch with the cutting edge of artificial intelligence development. The focus of prompt engineering had long since shifted to system constraints and tool design, rather than user input.
Zhao Wenyuan already knew how much of this task Glutinous Rice Ball could complete. What interested him now was how well the other models would do.
By now, the big screen had been divided into more than a dozen smaller sections, each showing a live feed from a sandbox environment. The models were independently breaking down the task and executing it.
Because these were sandbox environments, everyone, including the staff, could only see their visual output. They couldn't tell what programs were running inside. At this stage, it really wasn't obvious which models were stronger or weaker.
After watching for