Something no one talks about enough:
CPU cycles aren't free.
Memory reads and writes aren't free.
That shit takes power. A miniscule amount per instance, yes, but it adds up quick if it's from an app that everyone uses constantly. All that power comes from somewhere, and right now that's mostly fossil fuels.
When people complain about software bloat, it's not just a UX problem, it's an environmental one too.
I wish vector search as a thing would die, or return to the back seat and let traditional indexing things have the first shot.
"oh but vector search is great because it clusters similar concepts!"
Yeah you know what concepts are similar? Different names, different model numbers for things, things that are related to _but different from_ what I searched for.
If I search for "morgan", a result for "morton" is a bad result. If I search for "atari ste", a result for the atari falcon is bad.
It'd be cool if there was an approach of using "visual embeddings" applied to facial recognition. One could teach their personal agent a face by telling it to read the facial embedding and when encountering people the facial embeddings could be pulled from a regular vector DB instead of trying to train a new recognizer on the face. Probs easier to share data that way too.
This week's #Rust type of the day is `Result<Option<Result<Bytes>>>`
I don’t want a computer with adverts in the task bar or menu, on which I can’t run my own software freely and easily, or which I can’t repair myself.
I don’t want an eReader which snitches on my reading, or which can remove my books.
I don’t want to use a proprietary third party service and app for my dishwasher.
I don’t want a car which can be controlled remotely via someone else’s servers.
Hackers showed me (there's video) how a website vulnerability let them locate, unlock, honk the horn, start ignition of any of millions Kias in seconds, just by reading a car's license plate.
They found similar bugs for a dozen carmakers over the last two years.
https://www.wired.com/story/kia-web-vulnerability-vehicle-hack-track/
iroh is about creating connectivity to empower user agency. Reliable connectivity you can build on now and in the future.
https://www.iroh.computer/blog/smaller-is-better
TBH I don't think most people should be on algorithmic social media. Happy people aren't chronically online and these places are machines for brainworm infection and cortisol spiking.
I think "friend only" social timelines and read-only feeds of stuff you're interested in is okay though.
OFC I could never escape the draw of cyberspace, but I am a bit beyond salvation.
Advice for anybody starting a new Fediverse thing: The only real requirement in ActivityPub is that the content type is either application/activity+json or application/ld+json; profile="https://www.w3.org/ns/activitystreams". A lot of people have implemented this.
I think describing this as hazing against new projects might be appropriate.
Verilog to E. coli compiler (feat. Yosys) https://www.nature.com/articles/s41589-024-01730-1
For example, Chrome is working on shipping web APIs for LLM access. I'm planning to release something similar in @agregore in the next week or two.
https://github.com/explainers-by-googlers/prompt-api/blob/main/chrome-implementation-differences.md
I don't think cloud AI will fully go away but I think it'll make less and less sense for consumer facing use cases as the small models become more viable via better training and better hardware acceleration.
I think in the next couple years OS-shipped #LocalAI will replace the use of heavy cloud based #AI. Microsoft, Google, and soon Apple will be shipping devices with local LLMs and it'll be cheaper for applications to target those APIs rather than pay OpenAI or the such. This will also mean that we'll get into a sort of "browser wars" of model functionality gated by hardware vendors.
Occult Enby that's making local-first software with peer to peer protocols, mesh networks, and the web.
Exploring what a local-first cyberspace might look like in my spare time.