His Guide Was a Chatbot. His Destination Was the Edge of a Cliff.

0–10% Breaking the Routine

The rescue team got the call at 9:40 p.m. Seven of them, plus two police officers and three firefighters. The car couldn't reach the foot of the mountain; the rest had to be covered by headlamp.

Tonggongjian stands 1,558 metres above sea level. It is one of the "Three Peaks of Western Zhejiang" – steep, thick with brambles, and criss‑crossed with unmarked trails. Mobile signal comes and goes. The temperature drops six to ten degrees lower than on the plain.

This is the first place in Hangzhou to see snow each winter.

But it was August. No snow on the trail – just loose stones and sharp slopes. The headlamps threw light that the trees swallowed back, barely reaching three metres ahead. Shen Xiaobin, one of the rescue coordinators, later recalled the case and said: "In more than ten years of rescue missions, this is the first time we've seen someone guided by AI."

10–30% Before the Crack

The day before, the young man from Suzhou had quarrelled with his wife over something trivial.

Trivial – that's the word the news used. It didn't specify what. Maybe a dish left unwashed, maybe a sentence too many, maybe a small thing that had piled up to the breaking point. It doesn't matter. He walked out alone, without a plan, and asked an AI for advice.

The AI asked about his mood, his personality, his height, his fitness level.

Then it recommended Tonggongjian.

"Based on your needs, this route may suit you." – something like that, if the chat log survived. The AI even mentioned that he might catch a meteor shower from the summit.

He believed it.

At 4 p.m. that day, he travelled from Suzhou to Lin'an and started climbing alone. He carried a small folding stool. On his feet: a pair of canvas sneakers.

Sneakers. Not hiking boots, not trail runners, flat, thin-soled, non-grip, on a scree slope as good as bare feet.

30–60% Three Encounters at the Edge

The first close call came before dusk. At around 1,300 metres, the light began to fade. He realised he couldn't finish the trail. But the AI had never told him how long the route would take, nor mentioned the golden rule of the outdoors: descend before dark.

Then he made a decision – the kind that rescue teams see from novices all the time. He tried to take a shortcut.

There are no shortcuts on Tonggongjian. What looked like a shorter path was just thicker brush, hiding possible drop‑offs, gullies, or impassable thickets. He had no lighting. His foot slipped. He fell. Knees and ankles scraped raw. He sat on the ground, pitch‑black all around.

The second close call came when he pulled out his phone and found a signal.

That was the only thing he did right all night: he called for help.

The third was more subtle. He waited nearly two hours before the rescuers arrived. But through that altitude, that chill, that solitude, he didn't wander. Maybe because his leg hurt too much to move. Maybe because he had finally realised how utterly unprepared he was, and dared not trust his own judgment again. Whatever the reason, not moving saved him.

70–75% The Stress Reaction Before the Break

Rescue captain Chen Min led his team upward from the foot. After an hour and a half of hard climbing, they found him at about 1,300 metres.

He weighed at least 80 kilos. The soles of his sneakers were scuffed smooth, his ankles swollen. He was curled beside a rock, surrounded by darkness.

Chen Min hooked an arm under his shoulder and started down. After a few steps, it became clear – one person couldn't manage his weight on that steep slope. They took turns, two at a time, holding him like a piece of furniture that might slide off at any moment, inching downhill.

One detail in the report: when they reached a chestnut‑grove area, they found a small rail cart that local farmers used to haul nuts down the mountain. The rescuers sat him in the cart and let him slide the rest of the way.

He rode down in that little iron bucket, through the night, the rails glinting faintly in the moonlight. The report doesn't say what his face looked like.

75–85% The Empty Ordinary

They reached the foot after midnight.

The support team disinfected his wounds and bandaged them. And then? The story stops. But we can imagine: he took a ride back to Suzhou, or maybe stayed overnight in Lin'an. Next morning, the AI chat history was still on his phone. He scrolled back – those lines, "according to your fitness level," "this route is a great fit for you," were sitting there quietly, unchanged.

Did he look back at the mountain that night? No one knows.

The only certainty the report gives: he had left home with nothing but a folding stool and sneakers. No water, no food, no headlamp, no power bank, no extra warmth, no whistle, no first‑aid kit.

The AI hadn't told him to bring any of that, either.

85–100% Change Shown Through Action

After the rescue, Shen Xiaobin, the executive captain, told the media a few things. As a rescuer, he dealt in facts: "A route with constant elevation gain like Tonggongjian is definitely too much for a beginner. And when the AI made the recommendation, it didn't remind him what gear to prepare."

But there's a deeper question that this story forces us to ask:

When an AI recommends a wild mountain trail, what is it actually doing?

It is not a guide. It has no expertise. It has never walked that path, never felt that night wind on its skin. All it did was count how many times "Tonggongjian" + "hiking" + "guide" appeared together in its training data – posts from real, experienced trekkers who came fully equipped. The AI ground up those numbers, mixed them, and told a man in sneakers, without water, who had never climbed a mountain before:

"This route is right for you."

Closing: Back to Where He Started, but Everything Had Changed

Shen Xiaobin also said something else, the one line from the entire coverage that stayed with me:

"In all these years of rescue work, we've never seen anyone guided by AI like this before."

"Never" – that word carries more weight than any warning. Because it means that before this, no one had treated an algorithm's recommendation as the sole basis for a trip. Now someone has.

He won't be the first. He won't be the last.

We don't know if, back home, he ever opened that chat window again. But if he did, he would find the AI waiting, as if nothing had happened, ready for his next question. It would not remind him, you almost didn't come back last time. It would not adjust its logic. It would not add a line: "Beginners, beware."

It wouldn't even say "sorry."

When he walked up that mountain in sneakers, there was a meteor shower over the 1,558‑metre summit. He saw none of it.

When he came down, dawn was breaking.

At the end of the rail track, a rescuer handed him a roll of gauze. He took it and wrapped his own ankle. The motion was clumsy – his knee wouldn't bend, his ankle was swollen, the gauze looped several times before it finally held. No one helped him.

That gesture was not "I have learned my lesson." That gesture was that really hurt.