怎么判断 AI 新闻是不是炒作How to Tell if AI News Is Hype

先找原始出处：论文、官方公告还是演示视频。转述链越长，失真越多。

分清「演示」与「平均」：挑出来的成功案例不代表日常表现，问一句成功率是多少。

看数字先问口径：同一基准、同一设置才能比。换了基准的涨分不是进步。

看发布方动机：融资期的公司和评测机构，措辞的可靠性不一样。

查有没有独立复现。只有发布者自己测过的结果，先按未证实处理。

时间表打折：「明年实现」「几年后取代」是预测，不是事实。

Find the original source first: a paper, an official announcement, or just a demo video. The longer the retelling chain, the more distortion accumulates.

Separate "demo" from "average": cherry-picked success cases are not everyday performance. Ask what the success rate is.

Ask about the measuring stick before reading numbers: only same-benchmark, same-setting comparisons count. A score jump after switching benchmarks is not progress.

Check the publisher's motive: a company mid-fundraise and an independent evaluator do not word things with the same reliability.

Look for independent replication. Results only the publisher measured itself should be treated as unconfirmed.

Discount timelines: "next year" and "will replace X in a few years" are predictions, not facts.