meehai

with the amount of traction this has gotten... coming with a clear set of experiments even on arxiv paper would be of great help to showcase your improvements. And if they're easily reproducible, they could get integrated in the mainstream inference engines as well, as the main point here is compression with little degradation.

Actively doing this. It indeed forces me to think things through, organize thoughts and speak them out. I open paint/miro to draw. It's good practice.

and with little data (i.e. <10Mb), this matters much less than accessibility and easy understanding of the data using a simple text editor or jq in the terminal + some filters.

it's skills first and then money and hardware for scale

A more skilled person that understands all the underlying steps will always be more efficient in scaling up due to knowing where to allocate more.

basically... you always need the skills and the money is the fine tuning.

Can you append new columns to a file stored on disk without reading it all in mempey? Somehoe this is beyond parquet capabilities.

Yeah, but if you can do topologies based on latencies you may get some decent tradeoffs. For example with N=1M nodes each doing batch updates in a tree manner, i.e the all reduce is actually layered by latency between nodes.

lossy encycopledia that can also do some short-term memory (RAG) things.

Tbh, the web won the application platform mostly because it's a standard. Everybody knows html, css and a little JS.

On the other hand, for mobile apps, there is still a device-specific mentality.

Imagine web apps being built with a different flavor for all the major browsers...

I hope that the same level of standardization comes to mobile apps too with the option to use more device-specific features on top of the generic UI.

Mine is much more barebone:

- one single machine - nginx proxy - many services on the same machine; some are internal, some are supposed to be public, are all accessible via the web! - internal ones have a humongous large password for HTTP basic auth that I store in an external password manager (firefox built in one) - public ones are either public or have google oauth

I coded all of them from scratch as that's the point of what I'm doing with homelabbing. You want images? browsers can read them. Videos? Browsers can play them.

The hard part is the backend for me. The frontend is very much "90s html".

the last point got me.

How can you idiomatically do a read only request with complex filters? For me both PUT and POST are "writable" operations, while "GET" are assumed to be read only. However, if you need to encode the state of the UI (filters or whatnot), it's preferred to use JSON rather than query params (which have length limitations).

So ... how does one do it?