跳到正文
@EpochAIResearch· @EpochAIResearch · X·· 2026-08-28AI 评分29
AI 导读

不过,Sol 的表现仍远低于最顶尖的人类玩家,我们估计后者在最高难度设置(A20)下的胜率超过 50%。 据我们观察,这个 AI 经常出现失误并做出可疑选择,原因包括对游戏的战略理解不足、推理错误,以及视觉框架的一些问题。这类失误的一些例子:

正文

However, Sol still plays well below the best human players, who we estimate achieve >50% win rate on the highest difficulty setting (A20).

Anecdotally, we often saw the AI blunder and make questionable choices due to a combination of poor strategic understanding of the game, reasoning mistakes, and some issues with the vision harness. Some examples of these kinds of mistakes:

来源:@EpochAIResearch · x.com