easily said

i doubt the recent uptick in traffic is from “stealing data” for training but rather from agents scraping them for context, eg Edge Copilot, Google’s AI search, SearchGPT, etc.

poisoning the data will likely not help in this situation since there’s a human on the other side that will just do the same search again given unsatisfactory results. like how retries and timeouts can cause huge outages for web scale companies, poisoning search results will likely cause this type of traffic to increase and further increase the chances of DoS and higher bandwidth usage.

bruh i know people in their 40s making 6 figures that couldn’t read an error message if it would save ten generations of their family.

member when all the big cool web 2.0 companies had public facing APIs?

this is just combining existing data scraping tools with LLMs to create a pretty flimsy and superfluous product. they use the data to do what they say. if they wanted to scrape data on you they can already do that. all they get from this is your interest and maybe some other PII like your email address. the LLM is just incidental here. it’s honestly not even as bad privacy wise as a “hot or not” or personality quiz.

i’ve used Chezmoi for years now pretty successfully. works on my Mac and Linux machines. it probably could be made to work on Windows. i am transitioning to NixOS, but i’ll probably keep using it anyway, since i still have Macs for work (and because they’re great laptops don’t @ me). the only real downside is that it only works for the home folder, so i have to manually control stuff for /etc, but i generally prefer user configuration for most tools anyway.

i had messed around with Ansible for this in the past, but i didn’t really like it for this use case. it’s been a while tho so it’s hard to say why.

not to pile on, but you might also look at GNU Stow. i decided against it, but it’s there.

obligatory i s’pose: https://github.com/covercash2/dotfiles

most Zionists i’ve met are white Protestants, and most Jews i’ve met aren’t Zionists…

ah yeah. maybe less well known, but i had a dev kit from Qualcomm that came with Ubuntu

always? Android runs a linux kernel, and they support all kinds of embedded systems that run Linux.

nushell is excellent for dealing with structured data. it’s also great as a scripting language.

not sure what you mean by expensive. i run language models on my laptop that are pretty good at this type of task. and, yes, these models are infinitely easier and cheaper ultimately than trying to change the human proclivity for attention seeking behavior.

you’ve not seen the type of email chains i get at work. personally i think it should be illegal to respond-all to an email chain with hundreds of people with “Great job team!!! 🎉”. but it would be great to have a LM to read it near instantaneously for me to be like “oh yeah there was a product release and here’s a few relevant metrics”. doesn’t matter if it’s 100% in on every subtle detail, and a decent summary could tell me where or if i even should dig into details.

yeah i see that too. it seems like mostly a reactionary viewpoint. the reaction is understandable to a point since a lot of the “AI” features are half baked and forced on the user. to that point i don’t think GNOME etc should be scrambling to add copies of these features.

what i would love to see is more engagement around additional pieces of software that are supplemental. for example, i would love if i could install a daemon that indexes my notes and allows me to do semantic search. or something similar with my images.

the problems with AI features aren’t within the tech itself but in the surrounding politics. it’s become commonplace for “responsible” AI companies like OpenAI to not even produce papers around their tech (product announcement blogs that are vaguely scientific don’t count), much less source code, weights, and details on training data. and even when Meta releases their weights, they don’t specify their datasets. the rat race to see who can make a decent product with this amazing tech has made the whole industry a bunch of pearl clutching FOMO based tweakers. that sparks a comparison to blockchain, which is fair from the perspective of someone who hasn’t studied the tech or simply hasn’t seen a product that is relevant to them. but even those people will look at something fantastical like ChatGPT as if it’s pedestrian or unimpressive because when i asked it to write an implementation of the HTTP spec in the style of Fetty Wap it didn’t run perfectly the first time.

i guess the rare thing is the public commitment, but Apple has generally had a good track record for updates compared to its Android counterparts, who have previously failed to meet their goals or set laughable goals like 2 years.

this data is not the world

i think most ML researchers are aware that the data isn’t perfect, but, crucially, it exists in a digestible form.

i mean, i’ve worked in neural networks for embedded systems, and it’s definitely possible. i share you skepticism about overhead, but i’ll eat my shoes if it isn’t opt in

there are language models that are quite feasible to run locally for easier tasks like this. “local” rules out both ChatGPT and Co-pilot since those models are enormous. AI generally means machine learned neural networks these days, even if a pile of if-else used to pass in the past.

not sure how they’re going to handle low-resource machines, but as far as AI integrations go this one is rather tame

“we don’t know how” != “it’s not possible”

i think OpenAI more than anyone knows the challenges with scaling data and training. anyone working on AI knows the line: “a baby can learn to recognize elephants from a single instance”. reducing training data and time is fundamental to advancement. don’t get me wrong, it’s great to put numbers to these things. i just don’t think this paper is super groundbreaking or profound. a bit clickbaity and sensational for Computerphile

gotem!

seriously tho, you don’t think OpenAI is tracking this? architecural improvements and training strategies are developing all the time

i really want to like Nix.

gave it a shot a few years ago, but i felt like documentation and community support wasn’t really there yet. this was long before Nix surpassed Arch in terms of number of available packages. now people still complain about documentation, especially of the Nix language. i see a lot of package authors using it, and that kind of tempts me to start using at least the package manager. but a lot of packages don’t. the allure of GitOpsing my entire OS is very tempting, but then there’s been these rumors (now confirmed) of new forks, while Guix splintered off much earlier. for something that’s ostensibly supposed to be the most stable OS, that makes me nervous. it also seems to have some nontrivial overhead—building packages, retaining old packages, etc.

the pitch for Nix is really appealing, but with so much uncertainty it’s hard to pull the trigger on migrating anything. heck, if i could pull off some PoCs, i think my enterprise job might consider adopting it, but it’s a hard recommend for me today as it was 5 years ago.