Scroll scroll scroll…
“Strata informs your decisions — it doesn't make them for you.”
I hope you don’t kill someone.
How do we know it’s providing correct data?
The page states it’s powered by Claude with nothing I could see that relates to its performance, metrics, processes, testing, etc. for something expected to be used in a safety application.
Clicked the link - ok looks like it's for ski touring, because it mentions avalanche stuff?
Checked the route catalog - wait, there's barely any ski tours here and it's a bunch of random hikes from around the country. Why is avalanche risk highlighted when 99% of hikers are probably gonna stay home if there's a big snowstorm? Is this by chance your personal hiking history? Why these hikes in particular?
Are avalanches the only way someone can die outdoors? What about snake bites, heat exhaustion, drowning etc.?
I am not sure what the app actually does vs gaia, alltrails, nwac app. The site only has one screenshot. What does this app add?
A better plain title would be
Strata: An LLM App that Tells You If Your Hiking Route Today is Likely Safe.
Jokes aside, couple changes come to mind:
- pitch more carefully as "may indicate when danger may be elevated"
- in app, on calm days insist "while prelim scan found nothing anomalous, click these links to find the original forecasts"
As of today, with this tech for this use case, the warnings should be the lede. Otherwise you're even more likely to have a "'Full' Self Driving" situation where we're lured into a false sense of safety.
Avalanche risk and its management are unknown to me, but not to ski patrols, experienced backcountry trekkers, scientists, and national weather services of many countries, etc. I saw no evidence of trials to test to see how the technology performs against real conditions.
If you can’t trust this product’s recommendations within some margin of safety then it’s value proposition disappears.
Why not market it instead as an experiment and seek some way of collecting data to improve the process or develop a model fine tuned to the inputs?