tim-fan · HackerLangs

4 gün öncediscuss

Is anyone making LLM-in-a-box for emergency supply kits yet?

I feel that would be handy in all sorts of situations when networks are down.

tim-fan

3 ay öncediscuss

Hey me too, I also have a DIY inventory system, maybe yours is more developed. But I also had the concept of items for n arbitrarily nested containers.

I also track historical movements to see which items are never used.

Recently I've moved towards everything being stored in numbered bags, which are hung in order on a line for O(N) retrieval. For storage it tells you which bag to put it in, for retrieval it tells you where it is.

I'm thinking more and more the optimal system will have a physical as well as digital component.

Also, I feel this system would be great for shared workshops at work places and maker spaces etc. I was just rummaging through our lab at work today, there's so many parts in the lab no one would know about, if it was inventorised with a good integrated (AI?) search function the equipment could be much more useful/available.

https://github.com/tim-fan/hordor

tim-fan

4 ay öncediscuss

OK right, so they're at 170 million now, maybe ~1 life saved. I see it might be a bit early to be making claims about that.

Also in comparison the cases of disruption/blocking intersections/emergency-services do seem significant - seems plausible in the right circumstances you could lose lives.

Seems they need to 10x their miles before they can start making confident claims about lives saved.

tim-fan

4 ay öncediscuss

Can they just report directly how many human lives they have saved since beginning operation? (of course, within some error bars)

Maybe that's too much of a statistical stretch.

But would be a good to-the-point number to have on hand for some waymo debates.

"yes they caused some disruption in an intersection in so-and-so scenario, but on the other hand they saved X number of human lives last year"

tim-fan

6 ay öncediscuss

Here's the repo, I've just added a demo video to the readme:

https://github.com/tim-fan/hordor

I was learning Django when I wrote it, today you'd probably get further quicker vibe coding from scratch.

I have about 100 items in storage today, I intend to add more, would like to optimize the workflow as I scale up.

Going forward I'd like to add:

  * more optimized storage/retrieval flow. The overall goal for the project is to  minimize this friction, as far as possible
  * AI enrichment - generate descriptions, aid with search etc. I'd love to be able to query my storage "how can I connect this thing to this old speaker?" and the storage responds eg "you have this cable, this adaptor, plug that into this cable, etc"

I've seen a few related projects but can't find the links just now. There's some cool projects that store items in little trays each with an LED, when you request the item the LED blinks for rapid retrieval. The numbered bags I used are slower for retrieval but cheaper and easier to set up.

I do enjoy thinking about the different options and tradeoffs for cost and storage/retrieval time. Also tradeoffs between time and (physical) space.

edit: formatting

tim-fan

6 ay öncediscuss

I'm storing small items in numbered bags. I take a photo and have a web UI for easy searching. It also tracks which items are used frequently vs not used at all in years.

Bags are stored in numerical order for quick storage and retrieval.

With this you do your decluttering from the web interface: search for items that haven't moved in years, flag for removal.

For frequently used items the system doesn't make sense - the storage and retrieval overheads are too high. But it pays off for any item you might forget the location of, or forget if you have it at all.

I feel we're overdue to have these types of digital front ends over our household item storage.

tim-fan

6 ay öncediscuss

On the other hand, insurance costs for robotaxis should be lower if they are able to drive significantly safer.

Then the one I'm more interested / excited for: optimizing the fleet for the cargo. If most trips involve single passengers, then most cars can be small electric single seaters. This can further reduce insurance costs as well as fuel, maintenance and depreciation.

I'd hope that's enough to offset the price of the sensors, compute hardware, and engineers to maintain the system.

But yes paying back investors; not sure how long that would lead to elevated costs for riders.

tim-fan

7 ay öncediscuss

Not to be confused with Bonxai (Fast, hierarchical, sparse Voxel Grid)

https://github.com/facontidavide/Bonxai

Is there some connection between Voxel grids and Bonsai trees that I'm missing??

tim-fan

9 ay öncediscuss

Hey, cool to see!

I'm running a similar but smaller project (5k MAU), my oldest map is central London in 1561

https://onamap.me/maps/London1561/

I got into it because I was interested in the technical challenge of registering GPS to maps which are very warped compared to reality, like very old maps or illustrated tourist maps.

My home page is here for more: https://onamap.me/

I also came across this similar project a while ago:

https://www.verbeeld.be/2024/11/17/using-gps-in-the-year-156...

Good luck continuing to build out the project!

tim-fan

10 ay öncediscuss

I'm hoping for a self driving taxi + trains combo to maybe solve the problem.

For one a self driving taxi fleet could take up vastly less space - you'd no longer need one car per person, you'd need far fewer parking spots, most cars could be single or double seaters again taking less space and running more efficiently.

The space savings could be used to boost rail-based public transit options, which would see more adoption as self driving taxis make last-mile transport cheaper and easier. A bunch of positive feedback loops driving public transit adoption and improvements.

Result is cleaner and more efficient transport for all, and vast amounts of space returned from serving cars to serving people.

At least that's the dream!

comments