Had both 0.8b and 2b answer the following. "explain what a TCP three way handshake is and the exact sequence of flags exchanged" only 2b got it right. Actually find it mixed with offline docs and wikis a great addon.
CVE-security&privacy
mastodon
4.5.7
Posts
The qwen3.5:2b model actually runs reasonable on a pixal 9 via termux and ran its reasoning and full response through claude not bad also was pretty fast over all reasoning took time but not due to being slow just it second guessed some areas but referenced rfc texts its pretty solid over all for an offline llm on a phone 4b would not run due to memory ceiling but 2b runs and is pretty good overall worth keeping on my phone as an added resource for offline use.
A 9b model beat a 120b model and is cheaper to run as well those of you thinking hardware is everything heres an example of how its not
I actually love AI what I disslike is companys shoving it down my throat via OS integration and privacy invasive features. AI I control completely and can decide how and where its used I love.
I find for 99% of situations location permissions are not needed and ive basically never enabled the global location toggle ever even when using maps just type in addresses find where you are approxamitly via nearby buisnesses/landmarks and follow it on the map to your destination location toggle avoided.
Ever see a local, offline, private 14B model explain TCP connection establishment from the actual spec on a $200 machine?
The creater of the video is using AI to make this including voice but the content is legit and created correctly not just thrown together
THIS IS...LINUX!!!
https://www.youtube.com/watch?v=HRYTuFYcQOY
I think I have found a new hobby archiving.
If you argue threat models to me when talking privacy and security actions the day you have your identity stolen and banks drained and people using your insurrance and your in bankruptcy ill just sit with a smile and say "HOWS THAT THREAT MODELING WORKING FOR YA?
Why privacy and security is now so important screw threat models that no longer is relevant things are so different now.
https://cybernews.com/security/global-data-leak-exposes-billion-records/
By making my computer completely terminal based and running my phone the way I do I have everything setup to maximize goals of privacy security and learning.
Decided my laptop will be gui browserless removed firefox which was the only gui browser I had now im running 100% terminal Im serious on my CLI skills and the only browsers on my computer I want to use is terminal based like w3m or lynx
The idea of threat models is a way of saying if your nobody dont worry about privacy and security but if your somebody you shoould worry about it and I think threat modeling is bullshit and a weak ass excuse for saying its ok to have a shitty posture towards your privacy and security while claiming companys respect it and trying to hold them accountable you might as well say companys have 2 versions one for nobody becouse there threat model does not align and one for the somebodys.
Cli is super powerful!!!!
Curl to see if a site allows scraping. wget to mirror entire site for offline viewing and lynx or w3m to read everything that was mirrored and navigate the site offline really good for html heavy sites javascript does not render but that is a good thing.
Google is a big but small blimp in peoples privacy issue everything you do can be a privacy issue using your debit card is a privacy issue walking into a store with cameras at the enteremce "every store you walk into" is a privacy issue people hyper focus on google while basically ignoring every other. Javascript is a privacy and security nightmare and most allow it by default.
Useful tip-Curl sends minimal info when grabbing files and you can even strip info like user agent with -A and even mimic a browser a privacy oreinted way to grab files
We talk about the government collecting our data and privacy and the loss of ownership of things why not work on archiving our own personal data and maintain storage of info thats my new goal accumalate data in differnet forms. Hard and soft slowly just accumalate data weather pdfs that could be usefull or any other offline resource im currently storing kiwix/.zim files on sdcards as one step toward the accumalation of data
Firefox in 2010 ran perfectly with 2gb of ram. Today the same browser to display the same content html videas and such can easily consume 8 to 10gb. What has actually changed?
Look at that meta dropped a new little privacy killer and the ball keeps rolling.
Letting javascript just randomly run on your browser and not enabling it on a persite basis to me is a no go! I used to allow it to just run but not anymore theres to many issues on the internet to allow javascript free run and it is dangerous for ones privacy as well.
Why do americans actually care what russia bans such as facebook we have our own issues we cant even stop discord they just volantarily said now we need more info on you even thoe we cant protect it ,yet were worried about russia.
You can show some people DNA evidence prooving its a dog with pictures to go with it and some people will still say that its a cat lmfao.