Enhancing Route Search on Mountain Project with AI-Driven Tags

Hello!

I believe it would be fantastic if Mountain Project could provide a search feature based on route characteristics. As a trad climber, I want to hone my wide crack jamming technique, but finding wide cracks in my area is challenging. Currently, the only way to do this is by searching the forum. If climbing routes had tags like "off-width," it would be much easier to filter and find these routes.

Recent large language model (LLM) services like GPT make this task straightforward. By providing protection info to the LLM, it can return trad protection tags like "tiny-finger-size," "finger-size," "wide," and "very-wide." While I'm not suggesting Mountain Project should create a chatbot, here is a screenshot of my personal chatbot looking up wide crack routes in Leavenworth, Washington. Another useful scenario for me is filtering crack boulders to practice my crack techniques without belay partners, using tags like "crack-boulder". Refer to the second screenshot for an example.

The possibilities are endless. Imagine being able to search for routes with characteristics like "roof crack," "steep sport route," "easy access areas," "shade in the afternoon," or "sketchy top out/mantle." An LLM can achieve this with approximately 90% accuracy.

I'm working in AI industry. I can provide my scripts and explain how to implement this feature. Your consideration is highly appreciated.

Thank you!

LLM based AI is stupid garbage that drastically scales up mediocrity by destroying the environment.  This trend can't end soon enough.

/thread

Andrew Gram wrote:

LLM based AI is stupid garbage that drastically scales up mediocrity by destroying the environment.  This trend can't end soon enough.

Some of it. But vector search can be super useful and that's basically what he's suggesting.

It's true that the AI industry is increasing power consumption. However, it would be wasteful to dismiss all AI features entirely. Not all AI applications are energy-intensive. The tagging process I'm proposing is very light on LLM work. Even for the entire Mountain Project database, it would only take about an hour to complete. The value it adds in terms of improving search functionality is significant. While vector search is indeed a powerful tool, it can be expensive. That's why my proposal focuses on tagging, which is a more cost-effective solution.

I hope Mountain Project can consider this idea to enhance our user experience without a significant environmental or financial burden.

eventually I built it for myself :p

https://mtnpj.fly.dev/

Jumpei Hirono wrote:

eventually I built it for myself :p

https://mtnpj.fly.dev/

Very cool!!!  Thank you for doing this. 

Reading your earlier posts I was skeptical, but with the searches I've done so far your tool is extremely useful!

such a nice compliment. I will prepare an iOS app mainly for the offline usage. :)

Terry E wrote:

Very cool!!!  Thank you for doing this. 

Reading your earlier posts I was skeptical, but with the searches I've done so far your tool is extremely useful!