In general, I think speech as input/output is under-explored. In the emergency scenario, in a stressful environment, having an expert in your ear you can talk to should work much better than having a big manual book to look up specific cases.
That's handy for situations where you might not really understand what you need to search for. Any search system that can ask you clarifying questions is going to be a big improvement.
Or where you need to combine several steps together but you don't yet know what those steps are.
There's probably other technologies that could do that, requiring lower resources but they'll come with different trade-offs around configuration.
Just having a Raspberry PI, a offline copy wikipedia and a RAG enabled small LLM would be quite useful or at least entertaining if you have to go off grid.
The search engine is indeed the last missing component from a sovereign stack. But I think this could be solved locally with little cost. Instead of indexing content on the web we should be indexing sources themselves - where to look for X? - like forums, blogs, docs, feeds, and specialized search engines. We could collectively amass millions of these search stubs that can be used by local models to go and fetch fresh information from the source directly. This means separating the routing layer from the information layer, we don't need to keep information cached from the whole internet locally. The search stubs could fit in a few GB about same size with the local LLM. The cool thing is that sources change much slower than information itself, so the search stub database could be refreshed at a slower pace. We could combine a few million generic stubs with a few hundred personal stubs generated from our own activities. It is trivial to generate these stubs by piggy backing on frontier models.
That way your machine that, eg, normally plays video games or does AI work can support relief efforts by supporting emergency response IT. You don’t need to mothball the machine, just have an “emergency” boot USB than can run the services from your home generator.
You don’t even need to bring it with you: turn it on and leave it “best effort” at home, while you continue to use it via WAN.
But OK, let's assume that: The power is out, but you have a generator with so much fuel you can run a desktop just fine; Your neighborhood will somehow make a mesh network; Your neighbors need some already stored information and the best solution for that is texting a chatbot rather than a survival/emergency handbook or Wikipedia; Your mesh-network will also be good enough to match the time-sensitivity of the questions.
Under those assumption, which of these sounds better?
1. Buying an "LLM-in-a-box for emergency supply kits", which you deploy so that your neighbors can ask questions (text over the mesh) of the offline chatbot.
2. Buying a satellite internet transciever for your emergency supply kit, so that your neighbors can ask questions of a much better chatbot and communicate with human experts, their worried relatives, and coordinate with rescue/relief efforts...
I’m only out the cost of the drive, which is like $40 and doesn’t require anybody on the other side cooperate with me.
- - -
More broadly…
You call it unlikely mixes, but we see it all the time:
- people already have a computer for gaming or work
- people (ie, “preppers” like we’re discussing) buy a generator for emergencies
- local emergency response sets up mesh networking during disasters, both official and unofficial
Have you ever tried to use a handbook you’re not intimately familiar with during an emergency? It’s rough.
For personal preparedness, nothing replaces familiarity and practice — eg, weekend survival trips and reading your manual ahead of time.
But for providing information in a random lookup manner to unpracticed people who weren’t prepared? Yes, I think an LLM/chatbot is the practical way to operationalize all that information which you stored (eg, survival guides or machine manuals).
Also, it’s unlikely a general purpose chatbot would be superior at survival advice to one specialized for that purpose — and indeed, is likely to refuse your questions as “unsafe” or “criminal”.
At current prices you are also out about $4k for a Spark to actually run the inference on, if you want a full LLM in a low-power package.
In general, I'm not sure why one would want to pin your survival to an expensive, hallucination-prone data source, when an offline copy of wikipedia with a little vector search attached to a Raspberry Pi can fulfil the same role...
Sounds like the absolute worst time to rely on a crappy little model that will inevitably hallucinate.
Knowing humans? They'd probably take it by force and run it for themselves instead of providing light and heat to surgeons and water sterilizers...
/daily dose of cynism