Proposal: A LLM (large language model) built in to Windy
-
Why not introduce a large language model (LLM) trained on general knowledge plus a working manual of Windy that can help users?
Users could use a built-in LLM to learn how to use Windy, or learn more about weather-related stuff.
I feel as if a low-cost LLM (that may not be as good as ChatGPT or copilot) could still be potentially really useful.Understandably, a lot of people just don't understand that LLMs can be wildly inaccurate, and it can often falsify things, so there would be a need for a label, and possibly a topic summary done by the LLM + suggestion for users to do their own research without AI.
Some example questions/prompts:
"when will it likely be the coldest between the hours of 6pm and 8pm in the next 4 days?"
"How can it tell when a thunderstorm is likely?"
"Can you show me a map of wildfires in my area?"
"What is the best forecast model?"
"Where is the nearest place that has relatively low winds?"
"What are these cone icons below the wind speed labels, and how can I read them? (explain like I'm five)"
"What accessibility options are there, can you turn them on for me? I'm [visually impaired/blind/etc]"
"Can you help me understand what these isolines mean?"
"what websites do you recommend for me to read more about interesting weather events and topics?"understandably, there should probably be an "opt-in" for this, as a lot of people would not want to be bothered with it.
Would this affect profit margins too much (costs of running LLM on processing and power/electricity) and not be worth implementing at the moment? / Would users be willing to pay extra for this (e.g. subscription-only feature? or an optional additional +$0.50 a month to support Windy and have access to LLM, which could also potentially further limit people abusing it)
Would it be too difficult to safely/properly implement?
Would people use it as a general LLM and waste usage of the chatbot on non-Windy or non-weather related questions?
Would people even possibly abuse the LLM too much?
Would this addition be seen as more of a nuisance or bad addition rather than useful? -
@spencer-h I really like this idea but now that Windy as bought a large steak into Meteoblue I think they’ll use Meteoblues AI models in the future. 🤞🏼
