半仙下岗预警:AI 预测赢麻了?

街边"半仙"靠一张嘴，AI 靠上千亿参数——你猜谁准？国际预测大赛放榜，AI 夺了冠，人类高手集体靠边站。油价、疫情、数据中心批文，它算得头头是道，还敢拿真金白银下注，赚得比谁都欢。今天这篇，我们看看"预言家"这碗饭，往后还端不端得稳。

Artificial intelligence now beats some of the best human forecasters

AI 如今已能击败部分最顶尖的人类预测高手

Crystal balls give way to LLMs （水晶球退位，大语言模型登基）

On September 5th yet another domain of human intelligence fell to the artificial variety. For the first time an AI won the seasonal Metaculus Cup, a proving ground for forecasters of a wide variety of events. Not only that, other AIs took the second and fifth places, leaving third and fourth for humans.

Hundreds of entrants had predicted the outcomes of questions that would be resolved by September. Would an American state or EU country restrict data-centre development? Would the hantavirus outbreak affect at least five people who were not passengers on MV Hondius? How much would Brent crude cost? The participants were scored on the distance of their prediction from the true answer. Those that were closest for longest won the greatest number of points.

The human participants in the competition shared a $5,000 cash prize in proportion to the number of points that they picked up. But this sum is actually less than the fifth-placer had achieved in the real world. By betting on questions on which its AI thinks the market is wrong, FutureSearch has seen a 6% increase since June on the initial $100,000 value of its portfolio on Kalshi, a popular prediction market. Even that, though, is chump change compared with the performance of one of the developers behind Preseen who, despite his firm failing to make the top five in the competition, had turned $35 into $1.94m over seven months, the sixth-best return in Kalshi's history.

Computers have long been used to make forecasts in narrow fields, such as the weather. But AI forecasters, like those on Metaculus, are a different breed. Instead of being trained specifically on data regarding the questions they are predicting the answers to, they are built using the same large language models (LLMs) that underpin the rest of the AI boom.

These bots therefore read the news and make sense of data in much the same way that human forecasters manage—except they do so far more broadly and swiftly. This lets them explain the reasoning behind their forecasts explicitly, a useful trait when persuading decision-makers to take their bets seriously.

Soothsayers, from astrologers to racing tipsters, have been around for the whole of human history. The modern business of forecasting, however, has numbers attached to it, making it easier to weed out the charlatans and no-hopers.

By betting on questions on which its AI thinks the market is wrong, FutureSearch has seen a 6% increase since June on the initial $100,000 value of its portfolio on Kalshi, a popular prediction market.

📖 中文翻译：FutureSearch 专挑自家 AI 认为市场有误的问题下注，自6月以来，它在热门预测市场 Kalshi 上那笔10万美元的初始本钱已经涨了6%。

Directions: Write a letter of about 100 words to your supervisor, Professor Williams, apologizing for missing the group presentation of your research team last Friday. Your letter should include:
1) your apology and the reason for your absence,
2) what you have done to make up for it, and
3) your request for another chance to present in person. (10 points)

Dear Professor Williams,

I am writing to offer my sincere apologies for missing our group presentation last Friday morning. On my way to campus, I slipped on the wet stairs at the subway station and had to be taken to a nearby hospital for a minor ankle injury, which kept me from attending on time.

To make up for my absence, I have prepared a written summary of my part and shared it with my teammates, who kindly covered my slides. I would be most grateful if you could spare me ten minutes this week, so that I may present my section in person and answer any questions you may have.

Again, I apologize for any inconvenience my absence may have caused.

Yours sincerely,
Li Ming

✍️ 翻译题干

Cities that once built their streets around the car are quietly redesigning them around the bicycle. Over the past decade, protected bike lanes have spread from a handful of northern European capitals to cities across Asia and Latin America. The reasons are practical rather than romantic. Cycling infrastructure is far cheaper to build than metro lines, and a single lane can move more people per hour than a road packed with cars. Local businesses, once fearful of losing parking spaces, have often changed their minds: surveys in several cities found that customers arriving on foot or by bike spent more each month than those who drove. The obstacles that remain are political rather than technical. Drivers complain about lost lanes, and politicians who cancel cycling projects rarely pay a price at the polls. Yet the cities that have pressed ahead report cleaner air, quieter streets and healthier residents.

📖 参考译文

看完这篇，先别急着替天气预报员难过——AI 赢的是"算得出"的事。日子里的"算不出"，照样归人类管：它能算准明天堵不堵车，却算不出你明天会遇见谁、聊出什么火花。所以与其焦虑被替代，不如把自己活成那个"小样本外的意外"。当然，该下的功夫一样都少不了——毕竟连 AI 夺冠，背后也是成吨的数据一口一口喂出来的。下期见。