Full linux system on phone, available as a desktop experience through hdmi out. Samsung phones have Dex, Pixel 8 has desktop mode in beta. In terms of general use, UI is pretty much there (Dex is ahead of Pixel 8 beta, but likely full desktop mode is coming). Code-server works pretty well for running visual studio code in a browser window. Portable external monitor with a keyboard or a lapdock is much nicer to carry, serves as an extra battery, and you get the advantage of having internet connection through the phone when there is no wifi.
But because of how termux is structured, compiling stuff isn't always straighforward. The team does a good job of building out the packages, but for example, stuff like pip install numpy should just work natively (in termux you have to install it as a system package). There is Proot, but its very slow.
I would pay for a better Termux that wraps all the necessary stuff so I can just do everything like I do under linux.
I lose a bunch of things like Google Wallet and Android Auto if I go with Librem. Im sure I could get some notion of it to work, but I really don't want to take the time to set it up when my existing phone already runs linux.
I want a simple RAG. Not even the RAG part, just AI backed searching of a document. I put a CSV in there. I ask it a question. It pops me back the closest match. I don't need it to be very smart, but it should be better than fuzzy searches. Specifically it has to match a question from a different language, and it has to match questions asked in different contexts. I want users to ask it questions like "I don't like the books recommended here" and it should eventually match up with "What are your favorite books on Indian history?" inside the CSV. So it's not just keywords, semantics matter too.
I'm surprised it doesn't exist yet. Plenty of tools do this offline. There's online cloud ones just hosting the vector DB. Not much that just handles a search API.
But it has to do this at scale, for tens of thousands of users, not just one user like most sites do. I feel like Algolia should be able to do this easily but it's not on the list.
So does the CSV have multiple languages within it? Why a CSV verse various documents in a directory (like Google Drive?).
Just curious, I'm working on AskJack.io and the multiple users, find anything, no matter the context, is exactly what we're doing. But we haven't tried to tackle languages yet.
"So does the CSV have multiple languages within it?"
For context, it's for Indonesia. So you do have urban areas and expats who can mostly speak English. But English isn't taught in school. Arabic, Chinese, Dutch are popular too.
Bahasa Indonesia is the formal version, but most will be more familiar with the various languages. Most states will have their own dialects, at least 5 major dialects. The difference is not like American/English, but closer to Scots/English.
The challenge is that many people are comfortable with writing in their own dialect, even if they can read others. So we can't possibly use all the languages that could be input, but it's fine if it outputs in a somewhat different dialect. For the most part, sentence structures are similar and "bua" and "buah" are still recognizable to people with different dialects, likely to AI as well. But programmatic algorithms wouldn't handle something like "buwa" properly.
To answer the question, the CSV will mostly just have informal Bahasa Indonesia. English questions will miss sometimes, so we've added English as well for the ones that miss. More data is more effective.
"Why a CSV verse various documents in a directory (like Google Drive?)"
Could be anything really, but we tend to have a loose format for data & notes, and then have a script that strips out or combines from other sheets into something that's cleaner to read and search. PDF, XLS, etc is hard to get consistent. But something like JSON, XML, YAML are still acceptable as long as we can script it.
AskJack seems to be a bit low code though. We can build our own and have reasons to have more control over certain bits. So a B2B solution over the RAG part would be nice.
Not exactly "hair on fire," but I have dumped a bit of money into fixing my problem with music "dyslexia" (blindness?). Much more detail: https://hw.leftium.com/#/item/31689274
I enjoy social dancing, but it's tragic I can't connect with the music at more than a basic level. And it makes dancing with some (more musical) partners difficult/uncomfortable.
My hair on fire problem right now is i want to get a list of all companies and their company descriptions, headcount range and sector that match a single company name, sorted by relevancy. If you can create an API for that and charge per API call (say $100 for 5000 calls), i would be your first customer
A singular inbox for my favorite web articles, bookmarks, and Twitter bookmarks, and anything else I want to add (preferably via API) that requires no extra steps or manual imports. Readwise was close to this until Twitter shut off API access.
Eg: I save an article or bookmark something and it magically appears in my singular inbox.
https://notado.app has had support for Tweets and multiple other types of UGC (like comments! from everywhere!) for a long time now. I am the main developer and you can look at my comment history here for more details on the philosophy behind the project.
You can also check out some of my feeds of saved content[1] to get an idea of the breadth of support for UGC.
There are some articles available too if you don't mind longer reads.[2][3]
This is what I subscribe to currently (along with Readwise) but it doesn't save tweets unless I also pay for an IFTTT Pro subscription as well. Correct me if I'm wrong. I'm trying to reduce my online subscriptions, but ultimately, this may be the way I go if I can find the value in it.
For saving Tweets, I kick off an archival job from the command line to archive.today|is|ph and then bookmark the resulting link with Raindrop via API call. Not sure if that helps you, I hope that it does!
Compiling a tiny NIC test program that needs dpdk, cuda, and the oversized driver bundle. Seriously build me a binary that I can run on Ubuntu 18 or maybe 20 and I'll be happy. Bonus if you can tell me how to reproduce.
But because of how termux is structured, compiling stuff isn't always straighforward. The team does a good job of building out the packages, but for example, stuff like pip install numpy should just work natively (in termux you have to install it as a system package). There is Proot, but its very slow.
I would pay for a better Termux that wraps all the necessary stuff so I can just do everything like I do under linux.