It's been pretty eye opening watching Craig Wright (of bitcoin fakery fame) flooding out LLM generated 'academic' papers and even having some of them accepted.
He's toast if SSRN were to adopt a similar policy.
> Kias have a “Massachusetts mode” flag hidden behind a service menu (that needs a dealer code) that disables telematics at the owner’s request.
I would be very concerned that the flag just continues to submit your data but with a "telematics disabled" bit set on it. This is absolutely how location privacy is implemented in some devices. Moreover, even if it is effective it could be remotely reset including accidentally as part of an update.
> then the car will use your phone as an internet connection and send all the same telemetry data back to Toyota [...] so I exclusively use CarPlay via USB.
I would be concerned that a passenger connecting their phone to it while I was driving.
In other cars I've been successful picking up the relevant modules for peanuts from surplus/scrap then just desoldering the RF-active components (like bt radios, etc) and swapping them in. YMMV but if it doesn't work you're just out the cost of a junk part.
Even if some radio feature is benign its existence means that its hard to be confident that there isn't some other telemetry feature you missed. With no connectivity at all you don't need to worry that you missed something because you can monitor the car with a spectrum analyzer and observe its never transmitting.
Unfortunately in some newer cars you can't swap any modules without a dealer tool to pair the module to the car, presumably in a bid to prevent third parties from fixing the car (presumably preventing people from lobotomizing their surveillance isn't on their radar yet).
Using an AGPL violating mystery meat binary plugin that you run on your host, which potentially compromises any airgap you put around your printer (it attempts to connect to bambu servers, or did last time I checked it) and potentially your entire host.
> word choice in AI notes that might have a sinister connotation
Potentially sinister due to the biases of the model, as the model may have been trained using internet content that has a lot more fictional titillating evil overlord board meetings than the actual mind-numbing real thing. Training that included extremist anti-corporate dogma might even bias the language models towards hallucinating the worst possible misinterpretation.
I've seen whisper hallucinate whole legal arguments whole cloth when the AGC was broken in it and the audio went quiet-- so I think the language models in it are more than powerful enough to politically load a transcript.
Good practice should be to minimize any unnecessary stored records because ANY record just means more processing costs in discovery and god knows how much extra cost in litigation should it happen to have an unfavorable interpretation in light of some impossible to anticipate future litigation.
But if AI transcription must be used it would be might be prudent to save a copy of the original audio along with it.
Don't get me wrong, I'm with you here, but we are back to the days when we had to rent mainframe time for compiling programs. Not because of software limitations, but you just didn't have consumer grade hardware capable of running them.
This time, however it's even worse, because it'll be a really long time until either we get consumer GPUs with enough VRAM for full models or LLMs that fit in 16-32GB capable enough to compete with cloud providers.
I run locally qwen3.6 27b on my 3090 and it's really impressive for what it is, but it is still generations away from being capable of delivering a level of quality that we can confidently default to solo drive them on a daily basis.
That is an excellent idea, once we, the GPU-poor mice, figure out who is going to bell the SoTA training cat. Chinese models being banned is well within the realms of lobbied possibilities.
The real game would be to put a “nothing of interest here” prompt injection attack in the original series of prompts so a LLM parsing them later would ignore the attackers’ session.
> Vandalism can be reduced by excluding fare evaders because that's a class of people rather than a class of devices.
Just observing: People who don't own an iPhone or modern android are also, generally, of a class -- and probably one banks would prefer to not do business with for profitability reasons.
People who don't have spyware/lockinware for principled reasons are currently rare enough to not matter in this analysis-- though sure, they're probably customers the bank wants.
> Just observing: People who don't own an iPhone or modern android are also, generally, of a class -- and probably one banks would prefer to not do business with for profitability reasons.
I don't know about that. There are plenty of retirees who want nothing to do with this "modern technology" while still having large amounts of retirement savings that the bank very much wants at their institution.
Small (and for that matter large) business owners also have a tendency to have complicated financial situations and correspondingly want to deal with them using a computer screen rather than a phone, and that's another class of customers banks are certainly not interested in driving away.
Meanwhile I take it you're implying that the people who don't have a smartphone to do banking on are undesirable poors, but those are the people who do use a phone for banking, because bargain bin Android phones are available for ~$15 and that's the extent of what they can afford for an internet device.
Whereas the people using the likes of GrapheneOS might well be a small percentage of the customer base but they're still generally the class of customers the banks like, i.e. tech people with upper middle class financial situations.
He's toast if SSRN were to adopt a similar policy.
reply