The filtration capabilities available to most users is pretty robust; depending on what you use to interact with the Fediverse. I thinik it would be possible to filter out problematic bots, users and even whole domain sources with the right kind of software.
I’m going to be bold enough to say we don’t have as wide of an AI/LLM issue on the Fediverse as the other platforms will have.
I’m certain that if someone did collect data from the Fediverse; it would become a hot topic and it might not be enough data anyways as the Fediverse is not mainstream enough normally. So the data and language collected here might skew in a few imaginable ways that one might find undesirable for a general model of word frequencies.
Also the fact that people might not appreciate that data being collected. Let’s be real. It’s too soon for such a project to begin. The AI TREND MUST DIE as it currently lives and it’s corpse must be rotted away completely. Now, in internet time that may not be all that long…a few to several years…the memory of the internet can be short-lived at times. It must, however, fade from the public conscience into some obscurity first.
Once the technology no longer lies in greedy hands again; new development can begin anew.
Now we wait for someone to build an absolutely wonderful chat app on top of this wonderful bit of PoC code…
I genuinely hope someone does. Imagine what this could do if this was routed over Tor using Private Services.
Run this over that; and you’d have a bullet-proof text chat. Wrap a nice GUI client around all of that and you have a proper secure, anonymous messenger with no problems. With a little more build-out; you could even implement the Matrix protocol over this wire-line and basically have full inter-federation and moderation over a secure wire protocol; allowing for complete privacy and client integration.
TL;DR: Matrix over PQChat over Tor. Think about it. A Post-Quantum Dark-Matrix web.
Can it? Maybe. It’s not impossible; but it isn’t practical and most ISPs limit their shenanigans to grabbing your unencrypted DNS requests.
Will it? Probably no; aside from the previously mentioned DNS redirections; they’re not interested in most people’s packets, only in how many they deliver.
Should you care? I won’t tell you not to take precaution, but I do urge you to consider your threat model carefully and consider the tradeoffs. When Security & Privacy goes up, Convenience and Functionality WILL go down. Balance your needs. Don’t put yourself in a state of Privacy fatigue.
Are there easy fixes? Maybe. I think a VPN or using Tor would solve your concerns here anyways; it’s not required that your modem be running OSS that you can control. If you can achieve it; that’s still good for you; but it’s not something to be sweating if your modem isn’t capable and your invasive ISP is the only effective option.
It feels like this vulnerability isn’t notable for the majority of users who don’t typically include “Being compromised by a Nation-State-Level Actor.”
That being said; I do hope they get it fixed; and it looks like there’s already mitigations in place like protecting the authentication by another factor such as a PIN. That helps; for people who do have the rare threat model issue in play.
The complexity of the attack also seems clearly difficult to achieve in any time frame; and would require likely hundreds of man-hours of work to pull off.
If we assume they’re funded enough to park a van of specialty equipment close enough to you; steal your key and clone it; then return it before you notice…nothing you can do can defend against them.
Yeah this seems like a non-issue to me as well; the source material for the models is probably the cause of this bias.
I also don’t think there’s a lot of sources for this manner of speaking. Let’s also not forget that there’s oftentimes instructions given to the LLM that ask it to avoid certain topics which it will in fact do.
This is why I use SearXNG; locally hosted on a container. It collates and sorts out the crap from all the engines at once. I get a useful list of results normally. My personal instance is configured to shotgun several engines at once and uses Wolfram Alpha or Wikipedia for informational boxes over other engines; if those services present a result.
My experience:
Typically, using your own VPN should suffice. Depending on your situation you can do other things as well. If you are unable to download these tools on the school network in question; do not attempt to do so again. Use a public or other network connection elsewhere to obtain the tools you need to bypass their crap.
For example, NextDNS could be helpful. By running their client app; ( https://github.com/nextdns/nextdns/wiki/Windows ) you can make sure all your DNS requests are encrypted. Similarly you could simply set up a local DNS server that you point Windows at which can redirect those requests over DNS-Over-(HTTPS or TLS) to a DNS provider of your choosing.
As someone who owns a similar set; I can estimate you’re probably dealing with the upsampling issue due to an OS configuration issue. You should try listening to MP3s and FLACs at 44100hz sample rate for your comparison. Not 48000, not 96000.
In general I don’t believe you can tell any difference between MP3 and FLAC if you listen to the audio at the intended sample rate.
Meaning that @44100hz with 8 bit samples; you can’t tell.
Listening at higher sample rates with higher bits per sample; sure…there’s lots of room for unwanted and even audible error. Audio interpolation algorithms are not miracles, not smart, and not even close to being finely psychoacoustically tuned to your ears in most cases.
If you say you can hear a difference…you are lying or you are cheating by playing back the MP3 over an audio pipeline with a higher sample rate and bits per sample. Anyone could hear the difference when cheating like that. Human hearing can span all the way up to 128Khz; but oftentimes most people can’t notice a credible difference even at 96Khz.
But if you listen to a 44.1Khz signal via a 96Khz set of equipment; you’ll pick out exactly when the audio output shifts between being 96Khz and 44.1Khz.
This is how you can tell when audio is a recording at a lower sample rate. Most hardware is capable of outputting 96Khz so long as you don’t put older things in your audio chain (The pipeline from file to your ears, and yes this includes software and your operating system as well).
The problem usually arises when something is upsampled. Going from 44.1Khz to 96Khz is noticeable when you “Compress” the audio signal to boost apparent loudness. Most low-end equipment and unaware software will do this sort of operation automatically when upsampling your audio to make sure the process does not render your audio too quiet to hear. Your ears can hear frequencies being clipped or limited to a certain volume as well; which can also happen a lot to prevent certain issues. Because most people are unable to regulate this hidden software aspect of their playback chain; you can sometimes hear it.
Luckily if you spend some time with proper DSP software and/or hardware, you’ll be able to unmuddle/unmix these mistakes in your chain. It does take time and patience; and you’ll need a large blend of HQ audio (like FLACs or MQTTs) as well as your standard “downsampled” audio (like MP3s and other lossy tracks), and you’ll be able to tweak things so that everything sounds good.
Software packages like Viper4Windows or Viper4Android are good starting points and are often easy to figure out how to use and offer a very wide and diverse range of controls you can use to adjust the audio to your needs and liking.
Because everyone’s ears are different; there’s also plenty of tools that claim to adjust for your individual ears…and those can be helpful as well in chasing your perfect flat audio response curve and equalizing things to your preferences.
The CEO is oftentimes a company policymaker; I think it would be foolish to ignore that fact.
I’ve been boycotting C-f-a for at least 15 years now; and I don’t tell my friends or suggest that my family eat there either; except as an emergency uber last resort. The gas station (burritos/sushi/hot-dog-warmer) would be suggested first.
My current partner(s) know and respect my feelings for the company and they feel roughly the same anyways; and so we never eat there.
Such a system might be constructed for one’s own scraping needs by taking any one of the current frontend/backends and customizing that behavior such that it could mitigate issues or ingest/ignore data based on your own inputs as well; such that your model could be “riding along on a human surfboard with human guidance”